MNRAS 000, 1–21 (2018) Preprint 19 April 2018 Compiled using MNRAS LATEX style file v3.0

Snake in the Clouds: A new nearby dwarf galaxy in theMagellanic bridge∗

Sergey E. Koposov,1,2† Matthew G. Walker,1 Vasily Belokurov,2,3 Andrew R. Casey,4,5

Alex Geringer-Sameth,6 Dougal Mackey,7 Gary Da Costa,7 Denis Erkal8,Prashin Jethwa9, Mario Mateo,10, Edward W. Olszewski11 and John I. Bailey III121McWilliams Center for Cosmology, Carnegie Mellon University, 5000 Forbes Ave, 15213, USA2Institute of Astronomy, University of Cambridge, Madingley road, CB3 0HA, UK3Center for Computational Astrophysics, Flatiron Institute, 162 5th Avenue, New York, NY 10010, USA4School of Physics and Astronomy, Monash University, Clayton 3800, Victoria, Australia5Faculty of Information Technology, Monash University, Clayton 3800, Victoria, Australia6Astrophysics Group, Physics Department, Imperial College London, Prince Consort Rd, London SW7 2AZ, UK7Research School of Astronomy and Astrophysics, Australian National University, Canberra, ACT 2611, Australia8Department of Physics, University of Surrey, Guildford, GU2 7XH, UK9European Southern Observatory, Karl-Schwarzschild-Str. 2, 85748 Garching, Germany10Department of Astronomy, University of Michigan, 311 West Hall, 1085 S University Avenue, Ann Arbor, MI 48109, USA11Steward Observatory, The University of Arizona, 933 N. Cherry Avenue., Tucson, AZ 85721, USA12Leiden Observatory, Leiden University, Niels Bohrweg 2, 2333 CA Leiden, The Netherlands

Accepted XXX. Received YYY; in original form ZZZ

ABSTRACT

We report the discovery of a nearby dwarf galaxy in the constellation of Hydrus,between the Large and the Small Magellanic Clouds. Hydrus 1 is a mildy ellipticalultra-faint system with luminosity MV ∼ −4.7 and size ∼ 50 pc, located 28 kpc fromthe Sun and 24 kpc from the LMC. From spectroscopy of ∼ 30 member stars, wemeasure a velocity dispersion of 2.7 km s−1 and find tentative evidence for a radialvelocity gradient consistent with 3 km s−1 rotation. Hydrus 1’s velocity dispersion in-dicates that the system is dark matter dominated, but its dynamical mass-to-lightratio M/L ∼ 66 is significantly smaller than typical for ultra-faint dwarfs at similarluminosity. The kinematics and spatial position of Hydrus 1 make it a very plausi-ble member of the family of satellites brought into the Milky Way by the MagellanicClouds. While Hydrus 1’s proximity and well-measured kinematics make it a promis-ing target for dark matter annihilation searches, we find no evidence for significantgamma-ray emission from Hydrus 1. The new dwarf is a metal-poor galaxy with amean metallicity [Fe/H]=−2.5 and [Fe/H] spread of 0.4 dex, similar to other systemsof similar luminosity. Alpha-abundances of Hyi 1 members indicate that star-formationwas extended, lasting between 0.1 and 1 Gyr, with self-enrichment dominated by SNIa. The dwarf also hosts a highly carbon-enhanced extremely metal-poor star with[Fe/H]∼ −3.2 and [C/Fe] ∼ +3.0.

Key words: Magellanic Clouds – galaxies: dwarf – Galaxy: halo – Galaxy: globularclusters: general – Galaxies: Local Group – stars: general

1 INTRODUCTION

There is no solution to the “missing satellites” problem if itis defined simply as the mismatch between the total count

∗This paper includes data gathered with the 6.5 meter Mag-

ellan Telescopes located at Las Campanas Observatory, Chile.†E-mail: skoposov@cmu.edu

of the Milky Way dwarf galaxies and the predicted numberof dark matter halos capable of hosting a galaxy, since thereis no problem. Once the observed number of the Galacticsatellites is scaled up to correct for the effects of the detec-tion efficiency and the number of DM sub-halos is scaleddown in accordance with our understanding of the physicsof star-formation, little inconsistency remains (Benson et al.2002; Koposov et al. 2008; Tollerud et al. 2008; Koposov et

c© 2018 The Authors

arX

iv:1

804.

0643

0v1

[as

tro-

ph.G

A]

17

Apr

201

8

2 S. E. Koposov et al.

al. 2009; Jethwa et al. 2018). However, to make the theoryand the observations match requires that the vast major-ity of the Galactic satellites have extremely low luminositiesbut occupy sub-halos with significant virial masses. Only inthe past 15 years, thanks to the plethora of imaging surveyssuch as Sloan Digital Sky Survey (York et al. 2000), Pan-STARRS (Kaiser et al. 2002), VST ATLAS (Shanks et al.2015) and the development of automated search techniques(e.g. Willman et al. 2005a; Belokurov et al. 2007; Koposov etal. 2008; Walsh et al. 2009), the satellite census has been ex-panded to include systems with luminosities of L ∼ 100 Lestimated to inhabit DM halos of at least ∼ 107 − 109M(e.g. Simon & Geha 2007; Koposov et al. 2009; Jethwa et al.2018). Thus, the “missing satellites” problem is perhaps bet-ter viewed as a dramatic divergence between the sub-halomass function and the dwarf luminosity function (Klypin etal. 1999), with a pile-up of literally invisible objects withmass-to-light ratios of order 105 − 106ML

−1 .

The faintest of these ultra-faint dwarfs are only observ-able in and around the Milky Way, thus making the Galaxya unique place to provide crucial constraints on theories ofUniversal structure formation. However, as a result of theexpansion of the discoveries into the ultra-faint regime, oneproblem appears to have emerged. Since most searches havea limiting surface brightness of 30 – 32 mag arcsec−2, manyof the recently detected systems tend to have small physicalsizes, often only a few tens of parsecs. This makes the clas-sification into dwarf galaxies and globular clusters ambigu-ous based on the object’s size alone (Gilmore et al. 2007;Martin et al. 2008). Furthermore, the dynamical tests forthe presence of dark matter in the faint systems are oftenfraught with difficulties caused by low numbers of memberstars accessible for spectroscopy (Walker et al. 2009c), stel-lar binaries (Koposov et al. 2011) and instrument systemat-ics (Simon & Geha 2007). As a result, some low luminositysatellites have oscillated between the two classes (e.g. Be-lokurov et al. 2014; Laevens et al. 2014; Bonifacio et al. 2015;Weisz et al. 2016; Voggel et al. 2016) while the nature ofsome remains uncertain today (Willman et al. 2005b, 2011).To circumnavigate these displeasing ambiguities, additionaldiagnostics of “galaxyness” have been offered, including, inparticular, one relying on the presence of a spread of heavyelement abundances (see Willman & Strader 2012) indica-tive of a prolonged star formation activity.

Most recently, our view of the Milky Way satellite pop-ulation has become even more curious. In the past 3 years,data from the various observing campaigns utilizing theDark Energy Camera - such as DES, MagLiteS (Drlica-Wagner et al. 2016), and SLAMS (Jethwa et al. 2017) - haveyielded a stream of new satellite discoveries (Koposov et al.2015a; Bechtol et al. 2015; Kim et al. 2015; Kim & Jer-jen 2015; Martin et al. 2015; Drlica-Wagner et al. 2015b).Many of these have been uncovered thanks to the superbquality of deep DECam data that has facilitated detectionof fainter and more distant systems. However, many werebrought to light simply due to the expansion of the imagingsurveys into the previously barely explored Southern celes-tial hemisphere. Remarkably, the bulk of the Southern dis-coveries are situated near the Magellanic Clouds (Koposovet al. 2015a), which in hindsight seems obvious. Initially, asthe data started to come in, the satellite over-density sig-nal was barely significant. However, at present, the satellite

positions (Drlica-Wagner et al. 2016; Torrealba et al. 2018)and their kinematics (Koposov et al. 2015b; Walker et al.2016) appear to favor the hypothesis in which the MagellanicClouds (MCs) have brought a large group of their smallercompanions into the Milky Way (Deason et al. 2015; Jethwaet al. 2016). At the same time, thanks to a deeper and morecomplete view of the Magellanic area, supplied by the DE-Cam and Gaia data, we have learned a great deal about theClouds themselves and their interaction with each other (seeMackey et al. 2017; Belokurov et al. 2017).

Around the Clouds, the relatively-unexplored portionof the sky that has the greatest potential to inform our un-derstanding of both the satellites of the Magellanic satellitesand the Clouds’ life as a pair is the area directly betweenthe LMC and the SMC. This paper presents results from acontinuation of the survey of the Magellanic bridge region,first presented by Mackey et al. (2017). While this surveywas motivated primarily by the study of the stellar tidaldebris field around the Magellanic Clouds (see e.g. Mackeyet al. 2016; Belokurov & Koposov 2016), it also provides agreat resource to trawl for new MC satellites. Here we reportthe discovery and a detailed photometric, spectroscopic andchemical study of exactly that: a new dwarf galaxy in thebridge region of the Magellanic Clouds. The paper is struc-tured as follows. In Section 2 we describe the photometricdata from Dark Energy Camera. In Section 3 we discuss thesatellite properties that we infer from the photometric data.In Section 4 we discuss the spectroscopic data and model-ing of a new satellite. Section 5 provides a discussion of allthe chemical, dynamical and morphological properties of thenew system, possible connections to the Magellanic Cloudsand prospects for dark matter annihilation searches. TheSection 6 concludes the paper. The properties of the stellardebris field around the Magellanic Clouds are discussed in acompanion paper by Mackey et al (2018).

2 PHOTOMETRIC DATA

This work utilizes the photometric data from two DECamprograms, 2016A-0618 and 2017B-0906, focusing on the re-gion between the Magellanic Clouds. The first results basedon the 2016A-0618 imaging have been presented in Mackeyet al. (2017). The DECam is a large (570 mega-pixel) mo-saic camera boasting 62 science CCDs, giving a combinedfield-of-view of 3 square degrees (Flaugher et al. 2015). Thecamera is installed on the 4-m Blanco Telescope located atCerro-Tololo Inter-American Observatory (CTIO) in Chile.The camera is equipped with ugrizY filters and has beenused for the last 6 years to conduct the Dark Energy Survey(The Dark Energy Survey Collaboration 2005; Abbott et al.2018) as well as a number of individual programs and mini-surveys (Drlica-Wagner et al. 2016; Schlafly et al. 2018).

The data for the program 2017B-0906 - the primarydata source for this paper - were acquired on 8 and 9 Octo-ber 2017. We observed the area covering the side of the Mag-ellanic bridge region closest to the Small Magellanic Cloud.Images were taken in g and r filters with exposure timesranging from 70 to 225 s in the g band and from 60 to 110s in the r band, where the variation was due to changing lu-nar illumination during the night. On average, in the survey,

MNRAS 000, 1–21 (2018)

Hydrus 1 3

each sky location was covered by 3 individual exposures andthe total observed area is approximately 140 square degrees.

The initial data processing was carried out using thecommunity pipeline (Valdes et al. 2014) which provided uswith calibrated single exposure images that were correctedfor bias, flat-field, cross-talk, nonlinearity as well as beingWCS-aligned and remapped. The subsequent image pro-cessing was very similar to that described by Mackey et al.(2017). We first created image stacks in g and r bands fromall of the available data using the SWARP software (Bertin2010). The sources extracted from each CCD frame wereused to construct the Point Spread Function (PSF) by run-ning PSFex (Bertin 2011). Then we used a custom versionof the SExtractor code (Bertin & Arnouts 1996) in dou-ble image mode on stacked and individual images to obtainuncalibrated forced photometry PSF fluxes from individualimages. The fluxes were then photometrically calibrated andaveraged for sources with more than one observation. Thelists of detected g and r-band sources were cross-matched toproduce the final catalogue with a magnitude limit of ∼ 23in both g and r.

The photometric calibration was done using the APASSDR9 catalogue (Henden et al. 2012). Because APASS hassignificant spatially dependent systematic errors reaching upto 0.05-0.1 mag, we first rectified the APASS magnitudesusing Gaia (Gaia Collaboration et al. 2016a,b) and 2MASS(Skrutskie et al. 2006) data. To perform the APASS correc-tion, we assumed that the APASS g and r magnitude forstars with 0 < J −Ks < 0.75 could be estimated from theGaia G band measurement and the polynomial function ofthe 2MASS J−Ks colour. This allowed us to compute the gand r magnitude HEALPIX (Gorski et al. 2005) correctionmap for APASS with resolution of Nside = 128 (correspond-ing to ∼ 0.5 × 0.5 degrees). Finally we brought the mag-nitudes from the APASS photometric system into the DESDR1 system (Abbott et al. 2018) using simple color-terms.The recalibrated APASS magnitudes together with the in-formation from the overlapping DECam exposures were thenused to establish zero-points for each DECAM exposure aswell as the CCD-specific zero-point offsets (similarly to Ko-posov et al. (2015a)). A comparison of the final calibratedcatalogue with the small patches of the DES DR1 that over-lap with our footprint reveals that the remaining systematicerrors in the catalogue are at a level . 0.03 mag.

Throughout the paper we use extinction corrected mag-nitudes based on the Schlegel et al. (1998) extinction modelwith the coefficients from Schlafly & Finkbeiner (2011). Toseparate stars from galaxies in our catalogue we use the cri-terion based on the SPREAD_MODEL parameter measured bySextractor in the r band: |SPREAD MODEL| < 0.003(Desai et al. 2012).

In order to ensure a uniform calibration for both ofour datasets, 2016A-0618 and 2017B-0906, we reprocessedthe data from 2016 together with data from 2017 to obtaina continuous coverage of ∼ 380 square degrees across theinter-Cloud region. Figure 1 shows the density distributionof distant 20.5 < r < 22 0 < g−r < 0.3 main sequence turn-off (MSTO) stars selected from our two Magellanic Cloudprograms as well as the data from the Dark Energy SurveyData Release 1. The combined footprint of our programsis traced by a blue contour showing the targeted region ofsky. We remark that the data do not show any discontinu-

−25−20−15−10−50510LMC [deg]

−15

−10

−5

0

5

10

15

BM

C[d

eg]

0

5

10

15

20

25

30

35

40

Figure 1. The density of MSTO stars (20.5 < r < 22 and 0 < g−r < 0.3 in the Magellanic Clouds region. We include data from ournew DECam observing program, data from Mackey et al. (2017)

and data from the Dark Energy Survey DR1. The data from our

Magellanic bridge observations is surrounded by blue contour.The location of newly discovered dwarf galaxy in between Large

Magellanic Cloud (on the left) and Small Magellanic Cloud (on

the right) is marked by a red dashed circle. The coordinate systemused for the plot is a rotated spherical coordinate system such that

the equator is going through both LMC and SMC centers. The

pole of this coordinate system is αpole, δpole = 39.61, 15.49 andthe longitude zeropoint is at the location of LMC.

ity either between our 2016 and 2017 data or between ourprogram and the DES survey, thus confirming the absenceof significant systematic offsets in our photometry.

A stand-alone feature immediately apparent in the stel-lar density map in Figure 1 is an overdensity at L,B =(−13,−5). This overdensity in the Hydrus1 constellation isthe subject of this paper. In what follows we classify thesatellite as a dwarf galaxy (see Section 4) and therefore it isreferred to as Hydrus 1 or Hyi 1 in the rest of the text.

3 SATELLITE PROPERTIES

3.1 Stellar population

We start by analyzing the color-magnitude diagram (CMD)of the overdensity circled in red in Figure 1. The left panelof Figure 2 shows the background-subtracted Hess diagramof the central 10′ of the object. A clear main sequence (MS)together with a part of the red giant branch are both read-ily visible, confirming that the object is a bona-fide satellite.The Hess diagram of the overdensity is also clearly differentfrom the color-magnitude distribution of foreground starsaround it shown in the right panel of the Figure. To mea-sure the distance to the satellite and constrain its stellarpopulation properties, we perform a fit to the Hess diagramof the system within the inner 10′ by using two different sets

1 The Hydrus constellation is different from Hydra. Hydrus is a

mythical creature usually represented by a male water snake.

MNRAS 000, 1–21 (2018)

4 S. E. Koposov et al.

−0.5 0.0 0.5 1.0 1.5

g-r [mag]

14

15

16

17

18

19

20

21

22

23

r[m

ag]

PARSEC Age=8Gyr,[Fe/H]=-2.2MESA Age=12Gyr,[Fe/H]=-3.5

−0.5 0.0 0.5 1.0 1.5

g-r [mag]

Figure 2. Left panel: Background subtracted g-r, r Hess diagram

within 10′ of Hydrus 1. Stars within an annulus with radii of 20′

and 60′ were used for the background. Two theoretical isochronesshifted to a distance of 31 kpc from MESA and PARSEC libraries

are overplotted in blue and red respectively. Right panel: Hess di-

agram of background stars around Hydrus 1 (in 20′, 60′ annulus).

of theoretical isochrones. More precisely, we use both PAR-SEC isochrones (Bressan et al. 2012; Marigo et al. 2017)2

and MESA isochrones Dotter (2016); Choi et al. (2016);Paxton et al. (2011). The isochrones are populated accord-ing to the Chabrier Initial Mass Function (IMF) (Chabrier2003) with an additional magnitude-dependent photomet-ric scatter based on the uncertainties in the data. We thenevaluate the likelihood of the observed CMD over a grid ofages, metallicities and heliocentric distances. The resultingbest-fit MESA and PARSEC isochrones are overplotted inthe left panel of the Figure 2.

The best parameters for the MESA isochrone are: ageof 12 Gyr, iron abundance of [Fe/H] = −3.5 and distance of31 kpc. On the other hand, for the PARSEC isochrone weobtain 8 Gyr, [Fe/H] = −2.2 and distance of ∼ 31 kpc. Wefind that MESA isochrones fit the observed Hess diagramsomewhat better, possibly because the data are pushing thePARSEC library to its lower limit of [Fe/H]=−2.2. Due tolarge systematic differences in isochrone solutions we do nottry to assess the uncertainties in the age and metallicitesinferred, however the formal uncertainty on the heliocentricdistance for each isochrone set is around 2 kpc. While CMDfitting is an acceptable method of deriving distances it isalso known to suffer from inaccuracies associated with theage-metallicity-distance degeneracies as well as biases asso-ciated with the isochrone libraries used. These issues can bemitigated by using horizontal branch stars (both pulsatingand not) which can offer better accuracy in distance in mostcases. In fact in Figure 2 we can see ∼ 4 stars at r ∼ 18and g-r. 0 that are likely Hydrus’s Blue Horizontal Branch(BHB) stars (some of these were later spectroscopically con-firmed; see Section 4). This group of four likely BHB starscan be used to measure the distance by applying the fiducialrelation for the horizontal branch from Deason et al. (2011).As that relation was defined for the SDSS photometric sys-

2 We use the PARSEC v1.2S,COLIBRI PR16 isochrones down-

loaded from http://stev.oapd.inaf.it/cgi-bin/cmd in Febru-ary 2018

tem, we recalibrate it into the DES system. The resultingequation that we adopt is:

Mg = 0.444 + 0.106 c+ 0.704 c2 + 1.126 c3 + 0.794 c4

where c = (g − r)/0.3. With four BHB stars we derive thedistance modulus m−M = 17.20± 0.04 or distance of D =27.6±0.5 kpc, which is roughly consistent with the isochronefit result. The expected systematic uncertainty on the BHBmeasurement should be of the order of 10% (Fermani &Schonrich 2013).

To validate the distance measurement we have employedthe Magellanic RR Lyrae catalog from the OGLE-IV project(Udalski et al. 2015) which covers the location of Hydrus 1.Somewhat suprisingly, within twice the half-light radius ofHyi 1 we identified two RRab Lyrae stars, OGLE-SMC-RRLYR-6316 and OGLE-SMC-RRLYR-6325, with distancemoduli of 17.22 and 17.19 respectively3. With an averagedensity of 0.12 stars deg−2 for RR Lyrae in the distance mod-ulus range 17 < m −M < 17.4, the probability of havingtwo Galactic RR Lyrae within twice the half light radius ofHyi 1 is P ≈ 2×10−4. This suggests that both RR lyrae aremost likely genuine members of Hyi 1. The RR Lyrae fullycorroborate the distance estimate of m−M = 17.2 that weobtained based on BHB magnitudes.

With period and V-band amplitude combination of(0.67d, 0.8) and (0.73d, 0.56), the two RR Lyrae in Hyi 1are of Oosterhoff type II. This is in perfect agreement withthe recent discovery that the fraction of Oo I/Oo II typesdepends strongly on the stellar mass of the host (see e.g.Fiorentino et al. 2015; Belokurov et al. 2018), with higherluminosity objects (e.g. massive dwarf galaxies) containinga larger fraction of Oo I RRab stars. Extended down to thetypical luminosities of UFDs, this trend predicts that theRR Lyrae contents of objects like Hyi 1 should be almostentirely composed of Oo II type pulsators.

3.2 Spatial distribution

This section looks in detail at the spatial distribution ofstars inside the Hydrus 1 satellite. Figure 3 gives the posi-tions of the MSTO stars with 0 < g−r < 0.3, 20.5 < r < 22in 1 square degree region around the center of the satel-lite. We clearly see a high-contrast over-density with somehint of elongation approximately along the right ascensiondirection. In the Figure we also overplot the locations of theBHB stars and the RR-lyrae discussed in the previous sec-tion, thus confirming that they are clustered near the cen-ter of Hydrus 1. To describe the morphological propertiesof Hyi 1, we model the stellar distribution with a combina-tion of uniform background and flattened exponential profile(similar to Koposov et al. 2015a). We adopt standard priorson the model parameters: a Jeffrey’s prior on the exponentialscale length, uniform U(0, 1) priors on the object ellipticityand the fraction of stars belonging to Hydrus 1, a U(0, 180)prior on the position angle and an improper uniform prioron the position of the center of the object. We then sample

3 To compute distances to the RR Lyrae we use the period-

luminosity relations from Catelan et al. (2004), metallicity phase

relation from Smolec (2005) as prescribed by Pietrukowicz et al.(2015)

MNRAS 000, 1–21 (2018)

Hydrus 1 5

323436384042

α [deg]

−80.25

−80.00

−79.75

−79.50

−79.25

−79.00

−78.75

−78.50

δ[d

eg]

Figure 3. The distribution of MSTO (0 < g − r < 0.3, 20.5 <r < 22) stars in the vicinity of the Hydrus 1 satellite. The dashed

grey ellipse shows the best-fit shape of the satellite where the

major axis of the ellipse is twice the half-light radius. The arrowin the left corner shows the direction of the LMC’s proper motion

and the dashed line is parallel to the line connecting LMC andSMC. Red crosses show the locations of BHB candidate stars

with distance modulus within 0.1 magnitude of Hydrus 1, while

red star symbols identify the locations of RR Lyrae stars withdistance modulus within 0.1 magnitudes of Hydrus 1. We note

that both the RR Lyrae and BHB stars show clear concentrations

near the center of the object.

the posterior using a Markov Chain Monte-Carlo (MCMC)technique, specifically with the ensemble sampler (Goodman& Weare 2010) as implemented by Foreman-Mackey et al.(2013). The summary of our measurements is provided inTable 1. Here and throughout the rest of the paper, whenwe report the measurements and their associated uncertain-ties we use the mean and standard deviation in the case ofsymmetric Gaussian-like posteriors and the median and 15.8and 84.2 percentiles for asymmetric posteriors. In the Tablewe report both the major-axis half-light radius (i.e 1.67hwhere h is the exponential scale-length along the major axis)as well as the circularized half-light radius (i.e. multiplied by√

1− e).The angular half-light radius of the satellite is ∼ 6.6′

while its physical size, assuming a distance of 28 kpc, is ∼50 pc. These values suggest that Hyi 1 is a dwarf galaxyrather than a globular cluster (see e.g. Figure 6 of Torrealbaet al. 2016). We observe that the ellipticity of the system issignificantly different from zero: e = 1 − b

a∼ 0.2, confirm-

ing its slightly elongated visual appearance in Figure 3. Itis worth noting that the elongation of the system is approx-imately aligned with the Magellanic bridge and the propermotion of the Magellanic Clouds (shown as a red dashedline segment and a red arrow, respectively, in Fig. 3). Fornow, we refrain from speculation about whether this is amere coincidence or a tell-tale sign of a connection betweenHydrus 1 and the Magellanic Clouds. To verify whether theexponential model adopted for Hyi 1’s density profile pro-

0 5 10 15 20 25 30 35 40

D [arcmin]

0.1

0.2

0.5

1

2

Den

sity

[arc

min−

2]

Exp rhalf =6.62’

Figure 4. Azimuthally averaged density profile of Hydrus 1 turn-

off stars with 0 < g − r < 0.3, 20.5 < r < 22.25. The red curveshows our best fit exponential model.

vides an accurate description of the data, Figure 4 displaysthe azimuthally averaged 1-D density profile of the satellitetogether with the best-fit exponential model. Reassuringly,the model reproduces the observed profile well at all radii.

Equipped with a color-magnitude model and a spatialmodel of the satellite we can proceed to measure the absoluteluminosity of the system. We first compute the number ofstars in Hydrus 1 that are near our best-fit isochrone andbrighter than r = 21.75. This is approximately 1 magnitudeabove the limit of our photometric data and therefore shouldgive us an almost complete sample. Above this limit Hyi 1has an estimated 387 ± 28 stars, which, assuming the best-fit MESA isochrone at the distance from Section 3.1 and aChabrier IMF, gives a total stellar mass of ∼ 6000 M andan expected V-band absolute luminosity of −4.74 ± 0.084.We also checked that using the PARSEC isochrone yieldsthe same absolute magnitude within the uncertainty.

4 SPECTROSCOPY

In order better understand the nature of Hydrus 1, we ac-quired spectra of its brightest member stars.

4.1 Target selection

We selected potential members of the Hyi 1 system as thosethat lie spatially near the center of Hyi 1 and photometri-cally near the best-fitting isochrone measured in Section 3.1.The leftmost panel of Figure 5 shows the spatial distributionof all 235 spectroscopic targets (grey and black points), and

4 We remark that the total luminosity that we compute is the

expected luminosity given the total stellar mass and mass func-tion, rather than the estimate obtained by summing luminosities

of individual stars, which tends to suffer from large fluctuationsdue to poorly populated red giant branch.

MNRAS 000, 1–21 (2018)

6 S. E. Koposov et al.

Parameter Description Value Unit

αcen Right ascension of the center 37.389± 0.031 degree

δcen Declination the center −79.3089± 0.0045 degree

rhalf,maj Major axis half light radius 7.42+0.62−0.54 arcmin

rhalf,maj√

1− e Circularised half-light radius 6.64+0.46−0.43 arcmin

rhalf,maj√

1− e Circularised half-light radius 53.3± 3.6 pc

1− b/a Ellipticity 0.21+0.15−0.07

PA Positional angle of the major axis of stellar distribution 97± 14 degreem-M Distance modulus 17.20± 0.04

Distance Heliocentric distance 27.6± 0.5 kpc

MV Absolute Magnitude −4.71± 0.08 mag

σlos Line of sight velocity dispersion 2.69+0.51−0.43 km s−1

Vhel Mean Heliocentric radial velocity 80.4± 0.6 km s−1

VGSR Mean radial velocity in the Galactic rest frame5 −94.1± 0.6 km s−1

M(r<r1/2) Total dynamical mass inside half-light radius6 2.23+0.95−0.66105 M

M/L(r<r1/2) Mass-to-light ratio inside the half-light radius 6629−20 M/L

log10 M/L(r<r1/2/(M/L)) Logarithm of mass-to-light ratio inside the half-light radius 1.82± 0.16 dex

log10[J(0.5)/(GeV2cm−5)] Logarithm of the J factor 18.33+0.38−0.34 dex

dVheldα cos δcen

Radial velocity gradient along right ascension 24.4± 9.6 km s−1 deg−1

dVheldδ

Radial velocity gradient along declination 4.1± 9.2 km s−1 deg−1

< [Fe/H] > Mean spectroscopic metallicity −2.52± 0.09 dex

σ[Fe/H] Spectroscopic metallicity scatter 0.41± 0.08 dex

Table 1. Key photometric and kinematic properties of the Hydrus 1 dwarf galaxy. Reported measurements and uncertainties are either

means and standard deviations for symmetric posteriors, or medians and 15.8 and 84.2 percentiles in the case of asymmetric posteriors.

the second panel from the left shows the extinction correctedcolor-magnitude distribution of those same stars.

4.2 Spectroscopic data

On 12-13 Feb 2018, we used M2FS, a multi-object fiber spec-trograph at the Magellan Clay 6.5 m telescope, to acquirespectra for the selected targets. We used the same instru-ment configuration as in our previous M2FS observationsof ultra-faint dwarf galaxy candidates (Walker et al. 2015,2016)—specifically, we observed the spectral region 5130 -5190 A at resolution R ∼ 18, 000. We obtained 3 × 2700-second exposures (the middle exposure lost ∼ 300 secondsdue to technical problems) during conditions that were clear,with good seeing but high airmass (∼ 1.7 − 2.0). For cali-bration purposes, we also obtained high-S/N exposures ofthe solar spectrum during evening twilight, as well as Th-Ararc-lamp images immediately before and after the scienceexposures.

We processed and fit stellar-atmospheric models to allspectra following procedures identical to those described byWalker et al. (2015, 2016). For each star, we obtain si-multaneous estimates of line-of-sight velocity, effective tem-perature, surface gravity and metallicity. For all the pa-rameters we adopt uninformative flat priors with the ex-ception of the effective temperature, for which we use alog-normal prior based on the g − r color and spectro-scopic metallicity. Our prior takes the form of log Teff ∼N (S1(g− r) +A exp (B [Fe/H]), S2(g− r)) where A, B arefitted constants and S1(g−r) and S2(g−r) are cubic splines

5 Computed assuming the VLSR = 232.8 km s−1 from McMillan

(2017) and peculiar solar velocity from Schonrich et al. (2010)6 Obtained using the formula of Walker et al. (2009b)

with 20 equidistant knots in the −0.3 < g − r < 1.2 range.We calibrate this prior using the cross-match of the Dark En-ergy Survey DR1 photometric catalogue with the SDSS DR9spectroscopic catalogue, where the g − r colors were takenfrom DES and metallicities and effective temperatures frommeasurements by SDSS Spectroscopic Parameter Pipeline(Lee et al. 2008a,b; Allende Prieto et al. 2008). The me-dian parameter uncertainties for the spectra with signal tonoise larger than three are σv = 0.82 km s−1, σTeff = 80 K,σlogg = 0.19 dex and σFe/H = 0.13 dex. The velocity uncer-tainties include a systematic floor, added in quadrature, of0.2 km s−1, as measured from repeated observations of twi-light spectra.

Of the 235 spectroscopically observed stars, 117 havesignal-to-noise larger than 3 and 136 stars have radial ve-locity (RV) uncertainty less than 10 km s−1. As large RVuncertainties of 10 km s−1 and more occur with our high-resolution spectra only when the spectrum lacks discernablespectral features and/or has too low signal to noise, we ex-clude all the stars with σv > 10 km s−1 from further analy-sis. We show these stars in the second panel of Figure 5 byempty circles. As expected, most of these stars are faint andhave magnitudes r& 19.5, with the exception of two blue/hotstars that are likely BHBs that lack strong features withinthe observed wavelength range.

Table 2 presents spectroscopic parameter measurementsfor all the stars that either have signal-to-noise larger than3 or have radial velocity uncertainty less than 10 km s−1.We list measurements of the effective temperature, surfacegravity, metallicity and heliocentric line-of-sight velocitiy.We also provide in the table the g, r magnitudes and thelog-odds of being Hydrus 1 member star, based on the modeldescribed in the next Section.

The right three panels of Figure 5 summarize our spec-

MNRAS 000, 1–21 (2018)

Hydrus 1 7

ID α δ S/N Vhel σV Teff σT,eff log g σlog g [Fe/H] σ[Fe/H] lnPmPnm

g r

degree degree km s−1 km s−1 K K dex dex mag mag

2 37.69 -79.36 15.4 53.5 0.5 5723 64 4.53 0.12 -0.74 0.06 −64.1+17.3−24.7

18.45 17.92

3 37.70658 -79.35572 21.7 5.0 0.5 5353 35 4.49 0.08 -0.7 0.04 −415.8+119.8−169.4

17.86 17.25

4 37.74029 -79.35434 9.7 83.1 1.7 5521 82 1.4 0.46 -2.34 0.15 7.0+1.0−0.9

19.03 18.44

5 37.694 -79.34884 6.0 211.9 1.1 5904 114 1.16 0.84 -1.75 0.2 −1168.4+343.8−488.1

19.74 19.24

6 37.72267 -79.33451 18.1 -6.1 0.5 5437 33 4.81 0.1 -0.54 0.04 −545.6+157.1−224.0

18.21 17.63

7 37.69496 -79.32996 2.1 243.7 2.9 6269 240 3.35 0.86 -0.9 0.45 −1812.6+533.2−755.8

20.89 20.45

10 37.79812 -79.40667 4.7 83.9 1.5 5876 111 4.52 0.29 -1.58 0.19 1.3+1.6−3.0

19.81 19.29

11 37.83383 -79.38515 10.0 6.3 0.6 5792 91 3.18 0.19 -1.27 0.11 −396.6+116.7−164.5

18.84 18.33

12 37.75471 -79.349 3.7 187.4 1.5 5502 111 2.84 0.97 -1.88 0.27 −765.2+227.0−321.1

20.27 19.66

14 37.81337 -79.32914 2.2 85.7 6.5 6387 230 2.86 1.1 -1.0 0.47 −11.1+8.8−9.9

20.81 20.4

Table 2. Spectroscopic measurements for individual stars with either S/N>3 or radial velocity error less than 10 km s−1 in the Hydrus 1field. We include the internal ID of each star, right ascension, declination, signal to noise of the spectra, heliocentric radial velocity,

effective temperature, surface gravity together with their uncertainties. The log-odds ratios of being a Hydrus 1 member star vs being a

foreground contaminant from the chemo-dynamical model described in Sec. 4.3 and the g, r-band photometry from DECam observationsare also included. Only a subset of stars is shown in the printed version of the table, the full version of the table is available online.

troscopic results. Only stars with RV uncertainty smallerthan 10 km s−1 are shown. The third panel of the Figureshows the heliocentric velocity of observed stars vs distancefrom Hydrus’s center, and the fourth panel shows line-of-sight velocity vs spectroscopic metallicity. The fifth panelshows the 1-D distribution of velocities. These panels clearlyconfirm that Hydrus 1 is a real stellar system. A largeconcentration of stars belonging to Hyi 1 is prominent atheliocentric velocities of ∼80 km s−1 and low metallicities[Fe/H] < −2. This group of stars is kinematically cold andis noticeably more metal-poor than other stars in the fieldthat likely belong to the halo and/or MW disk. In total oursample seems to contain around 30 Hydrus 1 member stars.In the three rightmost panels of the Figure we also highlightthe stars that are confidently classified as giants/sub-giants,i.e. log g < 4, σlog g < 1 (shown in black rather than grey).This particular subsample of stars offers a particularly clearview of the Hydrus 1 RV peak at 80 km s−1, with littlecontamination at different heliocentric velocities. This is ex-pected given that stars belonging to Hydrus 1 should bemostly giant branch or horizontal branch stars and the con-tamination should mostly consist of thick disk dwarf starsand, to a much lesser extent, stellar halo giant stars at dis-tances of 30 kpc.

4.3 Chemo-Dynamical Modeling

As a next step we proceed to modeling of the observed kine-matic and chemical properties of Hyi 1 stars. To describethe distribution of velocities and chemical abundances inthe data, we adopt a standard mixture model with differentcomponents representing Hyi 1, MW halo and disk. We alsoutilise the fact that the sample of stars classified as giantsis much cleaner and should not be contaminated by thinand/or thick disk stars. Therefore we split our star sampleinto two subsamples: stars confidently classified as giants(log g < 4, σlog g < 1) and dwarfs (the rest).

A brief summary of the model is given below, where weprovide the distribution of velocity v and metallicity [Fe/H]conditional on the object type T , a latent categorical vari-able describing whether a given star is a halo star (T = h),disk star (T = d) or Hydrus 1 (object) star (T = o). Forall the components: halo, disk and Hydrus 1 we assume that

[Fe/H] abundances and velocities are independently Gaus-sian distributed. We also assume that radial velocity of Hy-drus 1 stars could have a linear gradient on the sky, param-eterised by two slopes Sα and Sδ along right ascension anddeclination respectively.

v|T = o, α, δ ∼ N (Vo +

Sα (α− αo) cos δ0 + Sδ (δ − δo), σo)v|T = h ∼ N (Vh, σh)

v|T = d ∼ N (Vd, σd)

[Fe/H]|T = o ∼ N ([Fe/H]o, σ[Fe/H],o)

[Fe/H]|T = h ∼ N ([Fe/H]h, σ[Fe/H],h)

[Fe/H]|T = d ∼ N ([Fe/H]d, σ[Fe/H],d)

Here the parameters Vh, Vd, Vo, σh, σd, σo refer to themean velocity and velocity dispersions of the halo, disk andHydrus 1 populations respectively, while [Fe/H]h, [Fe/H]d,[Fe/H]o, σ[Fe/H],h, σ[Fe/H],d, σ[Fe/H],o refer to the meansand dispersions in metallicity for the halo, disk and Hy-drus 1.

The last part of the model describes the distributionover types T for the samples of giants and dwarfs, condi-tional on angular distance bin from Hydrus 1’s center (Dbin;we use 5 distance bins). We assume that the sample of gi-ants has zero contamination from the MW disk and that theprobability of observing a Hydrus 1 member is a monotoni-cally declining function of distance. This reflects prior knowl-edge that the stellar density of Hyi 1 member stars withrespect to contamination decreases with radius, without as-suming a specific density law. This approach is preferred dueto the complexity of taking into account the fiber allocationefficiency (this is similar to the non-parametric approachadopted in the EM-algorithm by Walker et al. (2009a)).

T |Dbin = k, giant ∼ Cat((fo,g,k, 1− fo,g,k, 0))

T |Dbin = k, dwarf ∼ Cat((fo,w,k,

(1− fo,w,k) fh,d, (1− fo,w,k) fd,w))

The g and w subscripts in the parameters of the modelrefer to giant and dwarf samples respectively. For examplefo,g,k refers to the probability of observing a Hydrus 1 star

MNRAS 000, 1–21 (2018)

8 S. E. Koposov et al.

36.537.037.538.038.5

α [deg]

−79.5

−79.4

−79.3

−79.2

−79.1δ

[deg

]

65<Vhel<95 σV<10 [Fe/H]<-1.5σV >10

0.0 0.5

g-r [mag]

14

15

16

17

18

19

20

21

r[m

ag]

65<Vhel<95σV<10 [Fe/H]<-1.5σV >10

0.00 0.05 0.10 0.15

Distance [deg]

−200

−100

0

100

200

300

Vhe

l[k

m/s

]

log g<4 σlog g <1

−3 −2 −1 0

[Fe/H]

log g<4 σlog g <1

0 25Nstars /bin

Figure 5. Left panel: Spatial distribution of spectroscopically targeted stars. The center of Hyi 1 is shown a by red cross. Darker circles

indicate likely members, based on velocity and metallicity, while stars without a good velocity measurement are shown by empty gray

circles. Filled grey circles show all other stars. The dotted line shows the half-light ellipse. Second panel: Color-magnitude diagram ofspectroscopically targeted members. The meaning of the symbols is the same as in the left panel. We note that several horizontal branch

stars are marked as likely members; however, a few stars at the very edge of the horizontal branch avoid this distinction due to high

velocity uncertainties caused by lack of lines. A cross-hair marker indicates the carbon-enhanced metal poor star (Star 75) that is amember of Hydrus 1. Third panel: Heliocentric line-of-sight velocity vs angular distance from Hyi 1’s center, for all stars with good

velocity measurements (errors σv < 10 km s−1). Stars classified confidently as giants according to their spectra are shown by darkermarkers. Fourth panel: Heliocentric velocity vs spectroscopic metallicity for stars with good velocity measurements. The symbols used

are the same as in the previous panel. Rightmost panel: Velocity histogram for the giant-only subsample (black) and all the stars with

good velocity measurements (grey).

in the sample of giants in k-th distance bin and fd,w refersto the probability of observing a disk star vs halo star ina sample of dwarfs (we assume that this probability is notdependent on the distance bin).

The model described above assumes error-free ve-locities and abundances; however, since all our mea-surements have non-negligible errors, we introduce theseuncertainties into the model by assuming that all ob-served velocities and metallicities are normally distributedaround their true values: vobs ∼ N (v, σv,obs) and[Fe/H]obs∼ N ([Fe/H], σ[Fe/H],obs) where vobs, [Fe/H]obsare the observed velocities and abundances, σv,obs,σ[Fe/H],obs are the observational uncertainties and v and[Fe/H] are true error-free parameters of each star.

The priors on the parameters of the model are mostlyuninformative, Vh ∼ U(−300, 300), Vd ∼ U(−300, 300),Vo ∼ U(60, 100), σh ∼ U(50, 300), σd ∼ U(10, 300),σo ∼ U(0, 15), [Fe/H]h ∼ U(−5, 0), [Fe/H]d ∼ U(−1, 0),[Fe/H]o ∼ U(−5, 0), σ[Fe/H],h ∼ U(0, 2), σ[Fe/H],d ∼U(0, 1), σ[Fe/H],o ∼ U(0, 2), Sα ∼ U(−50, 50), Sδ ∼U(−50, 50), and U(0, 1) on all object type fractions fd,w,fo,g,k, fo,w,k.

This chemo-dynamical model is implemented in theprobabilistic language STAN (Carpenter et al. 2017) andis available in the supplementary materials for this paper.We run the model on all stars with velocity uncertaintiesless than 10 km s−1 (136 in total). We sampled the poste-rior using a Hamiltonian Monte-Carlo technique (Neal 2012)and No-U-Turn-Sampler (Hoffman & Gelman 2011). Afterrunning 24 parallel chains for 10000 iterations each, we en-sured that the resulting chains are in convergence using theGelman-Rubin diagnostic (Gelman & Rubin 1992) for eachparameter as well as inspecting the marginal 2-D and 1-Dposteriors, all of which show a single, smooth, well-behavedpeak. To check that our model provides a good descriptionof the data, we created replicated datasets from our best-fit

−3 −2 −1 0

[Fe/H]

−200

−100

0

100

200

300

Vhe

l[k

m/s

]

−3 −2 −1 0

[Fe/H]

logg<4

Figure 6. The demonstration of our chemo-dynamical model.Left panel: The 2-D density of stars in metallicity and radial ve-

locity space showing our 3-component Gaussian mixture model.

Right panel: The mock sample of 136 stars from our model, in-cluding both the giants log g < 4, σlog g < 1 (black points) and

dwarfs (grey points). The figure is meant to be very similar to thefourth panel of Figure 5. The observational uncertainties on both

abundances and velocities have not been added to the simulated

data points.

model. These are shown in Figure 6. The left panel of theFigure shows the overall smooth density in metallicity vs ve-locity space, as described by our model (assuming the sameratio between giant and dwarf subsamples as in our data),while the right panel shows an exact simulated replica ofour dataset with 136 stars. The replication seems to resem-ble the actual dataset shown in Figure 5, providing someassurance that our model provides an adequate descriptionof the data.

The numerical results of the chemo-dynamical modelfit, i.e. the values of parameters and their uncertainties rele-vant for the Hydrus 1 system, are provided in Table 1, while

MNRAS 000, 1–21 (2018)

Hydrus 1 9

Parameter Value Unit

Vhalo 142± 22 km s−1

σV,halo 134+17−14 km s−1

[Fe/H]halo −0.83± 0.13 dex

σ[Fe/H],halo 0.74± 0.1 dex

Vdisk 38.2± 7.6 km s−1

σV,disk 49.5+6.7−5.7 km s−1

[Fe/H]disk −0.52± 0.03 dex

σ[Fe/H],disk 0.17± 0.04 dex

Table 3. Measurements of the disk and halo parameters from the

chemo-dynamical model

Table 3 lists the parameter values for the contaminant com-ponents (disk and halo). Another output from our model isthe membership probability for each star. We express theseprobabilities the form of log-odds of being a member starlog(Pmemb/Pnonmber) and we include them as a column inTable 2. The uncertainties in those probabilities are obtainedafter marginalizing over all model parameters. We stressthat membership probabilities for individual stars are oftenhighly uncertain and model dependent; therefore, outputtingjust a single value of membership probability without uncer-tainty as it is common in the literature is meaningful onlyif model parameters (such as velocity dispersion or systemicvelocity) are known precisely (which is almost never true).

Now we briefly summarize the main measurements fromthe chemo-dynamical model, with more detailed discussionin Section 5. The systemic heliocentric line-of-sight velocityof Hydrus 1 is 80.4 ± 0.6 km s−1, which corresponds to aGalactic Standard of Rest velocity of −94.1 km s−1. Giventhat the angle between the line of sight of Hydrus 1 and theline connecting the Galactic center to Hydrus 1 is only 17degrees it is likely that the satellite is now on its way to-ward perigalacticon. The other key parameter in the chemo-dynamical model is the velocity dispersion of Hyi 1, which isresolved and measured with high accuracy: σv = 2.69+0.51

−0.43

km s−1. Assuming that the observed stellar velocities cleanlytrace Hyi 1’s gravitational potential, the total dynamicalmass within the projected half-light radius of Hyi 1 can thenbe estimated using the estimator of Walker et al. (2009b)M(< R1/2) ∼ 2.2 × 105 M. This value corresponds to amass-to-light ratio of 66+29

−20 solar units. Since mass-to-lightratio uncertainties are heavily asymmetric, we also computethe logarithm log10 M/L = 1.82±0.16, for which the uncer-tainties are close to Gaussian. As this ratio is clearly higherthan that of the stellar population alone, this confirms susp-cions based on morphology and structural parameters: Hy-drus 1 is a dwarf galaxy dominated by dark matter.

The mean metallicity and metallicity dispersion of Hyi 1are [Fe/H] = −2.52 ± 0.09 and 0.41 ± 0.08, respectively.The average metallicity is consistent with the metallic-ity of other ultra-faint dwarfs and agrees with the stellar-mass/metallicity relation traced by other dwarf galaxies(Kirby et al. 2013). We also note that the scatter in metal-licities within Hydrus 1 is resolved and is significantly largerthan zero, which is another indicator of the dwarf galaxynature of the object (see e.g. Willman & Strader 2012).

The best fit values for the line-of-sight velocity gradientin Hydrus 1 are Sα = dVhel

dα cos δcen= 24.4 ± 9.6 km s−1 deg−1,

−40−2002040dVel

dα cos δ0[km/s/deg]

−40

−20

0

20

40

dVel

dδ

[km

/s/d

eg]

Figure 7. 2-D posterior density of the spatial line-of-sight ve-

locity gradient as inferred from the M2FS kinematic data. Theellipses represent the posterior volume corresponding to 1, 2 and

3 σ. The blue wedge shows the range of direction of possible

velocity gradients if they were aligned along the major axis ofHydrus 1. The width of the wedge reflects the uncertainty on the

measurement of the positional angle of Hyi 1’s major axis.

and Sβ = dVheldδ

= 4.1±9.2 km s−1 deg−1. We notice that thegradient along right ascension is marginally different fromzero (at ∼ 2.5 sigma significance level). To investigate thisfurther, we show the 2-D posterior on the velocity gradi-ent in Figure 7. We also show there the wedge that wouldcorrespond to the radial velocity gradient along the majoraxis of the system (the width of the wedge represents ouruncertainty on the position angle of the major axis). Wenote that the marginal velocity gradient signal seems to bealigned with the major axis of Hydrus 1. We discuss theimplications of this result in more detail in Section 5.1.

4.4 Chemical abundances

As shown in the previous section, our spectroscopic datasetcontains around 30 members of Hydrus 1, many of whichare quite bright (r < 18) and have high-quality spectra.Therefore we synthesised spectra of 27 high signal-to-noise(S/N>5) and high probability (PHyi1/Pbg > 1) Hydrus 1stars in order to estimate their detailed chemical abun-dances. The observed wavelength region includes transitionsof Mg (the Mg Ib triplet), C (the C2 Swan band), Fe, and to alesser extent, Ti and Ni. We used the stellar parameters Teff ,log g, and [Fe/H] estimated from the previous Section, withmicroturbulence vt estimated using the empirical relation-ship from Kirby et al. (2008). We used the Castelli & Kurucz(2004) plane-parallel model photospheres, and compiled aline list with atomic transitions from the Gaia-ESO Survey(Heiter et al. 2015) and molecular transitions (primarily C2)

MNRAS 000, 1–21 (2018)

10 S. E. Koposov et al.

from Ram et al. (2014)7. We used the 2014 July version ofthe MOOG spectral synthesis program (Sneden 1973).

We masked regions with stellar absorption lines andused a spline function to continuum-normalise the obser-vations (Casey 2014). Continuum normalisation was rela-tively straightforward for all but one star, Star 75, which wefound to be strongly enhanced in carbon, producing signif-icant absorption at wavelengths bluer than 5165 A. Carbonabundances could only be estimated for Star 75, where weadopted a 12C/13C ratio of 3.5, and we report upper limitson [C/H] for most other members. Despite using a C2 molec-ular line list that was compiled specifically for accurate pre-dictions of carbon from the Swan C2 bands, we found thatthere were significant discrepancies between the predictedand observed wavelengths of the C2 feature near 5165 A inStar 75. Although the S/N in this spectrum is relativelylow (S/N ≈ 6 pixel−1), the predicted ‘saw-tooth’ spectralfeatures seem to almost grow out of phase with the obser-vations (Figure 8), even after allowing for a small residualradial velocity shift to fit the observations. Adopting differ-ent isotopic fractions of 12C/13C did not resolve the issue.Discrepancies in wavelength position of this bandhead havebeen noted in other works (Brooke et al. 2013; Ram et al.2014), but not to the extent that seems apparent in Star 75.On the other hand, the low S/N ratio does make it difficultto properly quantify the extent of the wavelength discrepan-cies. The estimated abundance ratio of [C/Fe] = +3 seems tobe largely independent of these wavelength offsets, becauseif we convolve the observations and data with a Gaussiankernel to smooth out the spectral features and produce anear-continuous opacity blueward of 5165 A that is due tothe C2 Swan band, the model and observations fit excellentlywith [C/Fe] = +3. The molecular features, and the level ofnear-continuous opacity when smoothed, is extremely sensi-tive to the level of [C/H], so the estimated abundance ratioof [C/H] seems to be largely insensitive to the differences inthe wavelengths of the C2 bandhead.

We were able to estimate [Mg/H] abundances for allHydrus 1 members. The Ti and Ni atomic lines in this wave-length region are relatively weak, only allowing for one mea-surement of [Ni/H] = −2.15± 0.11 for Star 154, and upperlimits of [Ni/H] < −2.68 for Star 165, and [Ti/H] < −2.46for Star 52. Although these limits are not very informa-tive, we include them here for completeness. We were ableto estimate Fe abundances by spectral synthesis for moststars, except for those with very low S/N ratios (e.g., S/N≈ 5 pixel−1) or very hot stars (Teff ≈ 8000 K). These esti-mates are in very good agreement with those estimated bytemplate fitting in Section 4.2: well within the estimated un-certainties of either method. Uncertainties in chemical abun-dances were estimated during the fitting process by project-ing the derivatives in the synthesised spectra with the uncer-tainties in the continuum-normalised fluxes. The estimatedabundances, uncertainties, and upper limits, are given in Ta-ble 4. We discuss these measurements in the next section ofthe paper.

7 Available in MOOG format from http://www.as.utexas.edu/

~chris/lab.html

Star ID [C/H] [Mg/H] [Fe/H]

4 < −0.32 −2.64± 0.11 −2.42± 0.10

26 < −0.15 −3.04± 0.15 −2.15± 0.10

30 < −0.74 −2.27± 0.10 −1.74± 0.1034 < −0.64 −2.72± 0.12 −2.94± 0.10

49 < −0.82 −2.38± 0.10 −2.97± 0.10

52 < −1.13 −1.78± 0.10 −2.38± 0.1061 < −0.43 −2.07± 0.16 −3.08

68 < −1.03 −2.40± 0.10 −2.67± 0.10

69 < −0.71 −2.26± 0.11 −2.61± 0.1072 < −0.74 −2.53± 0.11 −2.80± 0.14

75 −0.18± 0.20 −2.77± 0.14 −3.1877 < −1.06 −1.01± 0.10 −1.60± 0.10

102 < 0.98 −2.61± 0.13 −3.17± 0.18

119 < −0.47 −2.43± 0.12 −2.87± 0.10139 −2.13± 0.10 −2.26

149 < 1.82 −0.70± 0.12 −1.40

150 < −1.28 −2.33± 0.10 −2.52± 0.10151 −2.28± 0.34 −2.31

154 < −1.43 −2.34± 0.10 −2.55± 0.10

165 < −1.58 −2.55± 0.10 −2.85± 0.10182 < −1.01 −2.57± 0.10 −3.01± 0.10

188 −2.26± 0.29 −2.23± 0.11

209 < −0.12 −2.36± 0.14 −3.12± 0.13221 −1.84± 0.10 −2.21

222 < 0.07 −2.24± 0.23 −2.68± 0.12225 < −0.71 −2.59± 0.10 −2.54± 0.10

Table 4. Chemical abundances and upper limits estimated via

spectral synthesis (Section 4.4) for high probability members of

Hyi 1.

5 DISCUSSION

In this section we discuss and summarize the key resultsfrom the photometric, chemical and dynamical analyses ofHydrus 1 presented in Sections 3 and 4. In particular wefocus on possible radial velocity gradient of the dwarf, itsdark matter content, chemical properties, connection to theMagellanic Clouds and dark matter annihilation searches.

5.1 Radial velocity gradient

The radial velocity gradient that we detect in Hyi 1 is onlymarginally significant, at 2.5 σ, and seems to be aligned withthe major axis of the system. If we assume that it is not astatistical fluctuation, possible causes of this gradient wouldinclude satellite rotation, perspective rotation and tidal dis-ruption. Perspective rotation (Kaplinghat & Strigari 2008;Walker et al. 2008) is caused by projection effects from tan-gential motion of an object with large angular size. Howeveras Hyi 1’s size is only ∼ 0.1 on the sky, the observed gra-dient of ∼ 24 km s−1 deg−1 would imply an unrealisticallyhigh tangential velocity (& 1000 km s−1), and is thereforean unlikely cause of the observed gradient. Another possibleexplanation is tidal disruption of the satellite. This has beenclaimed to be observed in a few MW satellites (Aden et al.2009; Collins et al. 2017), albeit at low levels of significance.We note however that the observed gradient is much largerthan expected from a simple velocity gradient along the or-bit in the MW gravitational potential. Therefore, in orderto explain the gradient in terms of disruption, significant

MNRAS 000, 1–21 (2018)

Hydrus 1 11

5150 5160 5170 5180

Wavelength [A]

0.0

0.5

1.0

Nor

mal

ized

flu

x

[C/Fe] = 3.00± 0.20

Figure 8. Spectrum of the carbon enhanced extremely metal-poor (CEMP) Hydrus 1 star (star 75) with spectral synthesis modeloverplotted in red. Gray shading shows the observational uncertainties in the measured spectrum and red shading shows the uncertainties

associated with the spectral synthesis parameters. The spectrum shows a strong carbon absorption band and low metallicity indicative

of CEMP stars.

fine-tuning of the orbit would be required (Kupper et al.2010; Martin & Jin 2010; Kupper et al. 2017). Furthermore,the low ellipticity of Hyi 1 and the apparent regularity ofthe satellite and its density profile argue against this pos-sibility. Finally, the velocity gradient in the system couldbe explained by satellite rotation. Assuming the measuredradial velocity gradient along the major axis correspondsto solid-body rotation, we estimate a rotation velocity of3 km s−1 at the half-light radius, which is comparable to thevelocity dispersion of the system. To illustrate the plausi-bility of this interpretation, we show in Figure 9 the radialvelocity of possible Hydrus 1 members (selected based onmetallicities only) as a function of position angle of the sys-tem, with the curve showing the line-of-sight velocity modelevaluated at the half-light radius (essentially interpreted assatellite rotation). We note that the model seems to fit thedata well, matching the observed RV changes as a functionof positional angle. If the rotation hypothesis is a correctinterpretation of the data, that would make Hydrus 1 thefirst ultra-faint dwarf with significant rotation. Why wouldbe Hydrus 1 be an exception? It is difficult to say for sure,but it is possible that at least some of the dwarf satellitesof the MW had disk-like morphology before accretion, andwere then transformed into spheroidal galaxies by a processknown as tidal stirring (Mayer et al. 2001; Kazantzidis etal. 2011). In this case some of the satellites could still bein the transformation phase with significant rotational sup-port. That would mean that Hydrus 1 became a satellitevery recently.

However, due to the low statistical significance of therotation signal, we urge caution and emphasise the need formore data to confirm or refute it.

5.2 Nature of Hydrus 1

While the nature of Hydrus 1 as a dwarf galaxy is establishedby its size, dark matter content and internal metallicity dis-tribution, in this Section we put Hyi 1 in context with othersystems of similar structural and chemo-dynamical proper-ties.

With a circularised effective radius of ∼ 50 pc and lumi-

0 100 200 300 400 500 600 700

PA [deg]

60

65

70

75

80

85

90

95

100

Vhe

l[k

m/s

]

Figure 9. Line-of-sight velocities of Hydrus 1 stars vs position

angle. Only stars with velocity uncertainty less than 5 km s−1 and[Fe/H]<-1 are shown. All the data points have been plotted twice,once with the original position angles, and once with position an-gles offset by 360 degrees. The red curve shows the radial velocity

change according to the best fit velocity gradient evaluated at

half-light radius (corresponding to ∼ 3 km s−1 rotation velocity).The shaded region around the curve has the half-width of the best

fit velocity dispersion of the system.

nosity ofMV ∼ −4.7, Hydrus 1 sits in an ambiguous betweendwarf galaxies and extended globular clusters (GCs) in theluminosity-size space. Figure 10 shows all known MW satel-lites (dSphs and GCs) together with Hydrus 1. The objectssimilar in size and luminosity to Hydrus 1 are the Hydra IIsatellite (Martin et al. 2015) with a luminosity ofMV = −4.8and size rh = 68 pc, and the Sagittarius II satellite (Laevenset al. 2015) with a luminosity of MV = −5.2 and half-lightradius of ∼ 37 pc. The Crater 1 (or Laevens 1) satellite (Be-lokurov et al. 2014; Laevens et al. 2014) has similar luminos-ity (MV = −5.3) to Hydrus 1, but significantly smaller size(∼ 20 pc). The classification of Hydra II is somewhat unclear

MNRAS 000, 1–21 (2018)

12 S. E. Koposov et al.

1 10 100 1000rh [pc]

−14

−12

−10

−8

−6

−4

−2

0

2

MV

[mag

]

Cra 1

Sgr II Hydra II

Hydrus 1

MW dSphMW GCExtragalacticM31 dSph

Figure 10. Distribution of sizes and luminosities of various faint

stellar systems. Hydrus 1 is shown by a red star. MW/M31 dwarfspheroidal galaxies are shown by black filled/empty circles, re-

spectively. Large grey circles show MW globular clusters, while

small grey circles are showing extra-galactic compact stellar sys-tems (< 100 pc) from Brodie et al. (2011). Data are taken from

Figure 6 of Torrealba et al. (2018) and include the most recent

dwarf galaxy and globular cluster discoveries.

−14−12−10−8−6−4−20MV [mag]

1

10

100

1000

10000

M/L

(<r 1/

2)[

M

/L

Hydrus 1

Crater 2

MW dSphM31 dSph

Figure 11. Dynamical mass-to-light ratio as a function of lumi-

nosity for dwarf satellites of the MW and M31. Hyi 1 is marked bya red circle. Data come from the 2015 version of the dwarf galaxy

database by McConnachie (2012) as well as additional data by

Caldwell et al. (2017); Li et al. (2018).

because its velocity dispersion is only marginally resolved;however, (Kirby et al. 2015) argue that its chemical abun-dances favor a dwarf galaxy interpretation. Sagittarius IIdoes not have any velocity dispersion measurement. Finally,the Crater 1 satellite currently seems to be classified as anextended globular cluster based on the star-formation his-tory and unresolved velocity dispersion (Weisz et al. 2016;Voggel et al. 2016). We note however that the mass-to-light

ratio of Crater 1 from Voggel et al. (2016) M/L = 8.52+28.0−6.5

is formally consistent with both the absence of dark matterand a significant amount of dark matter.

While the mass-to-light ratio measurement oflogM/L = 1.82 ± 0.16 in Hyi 1 unambiguously indi-cates the presence of dark matter, it is worth comparingHyi 1 to other dwarfs of similar luminosity. Figure 11 showsmass-to-light ratio as a function of luminosity for MW andM31 satellite dwarf galaxies; Hydrus 1 is identified by a redmarker. We note that Hyi 1 seems to be an outlier, possess-ing a mass-to-light ratio well below (by a factor of 3) othergalaxies with similar luminosities. What makes Hydrus 1such an outlier? One possibility is that some of the faintobjects like Hydra II, Crater 1 and Hydrus 1 and others thatoccupy the area in size-luminosity space between globularclusters and dwarf galaxies could be intermediate-typeobjects with small but nonzero dark matter content andmass-to-light ratios smaller than other typical ultra-faintdwarfs. The origin of these systems could then be relatedeither to tidal disruption after accretion (Penarrubia etal. 2008) or to their initial formation (Ricotti et al. 2016).Testing this hypothesis would be quite difficult, however, asthe known objects near the dwarf galaxy/globular clusterinterface have small velocity dispersions and we lack highenough quality data on these systems to rule out a smoothtransition from dark matter dominated objects to darkmatter-free globular clusters.

Another possible explanation for the apparent low mass-to-light ratio of Hyi 1 could be related to our tentativemeasurement of the line-of-sight velocity gradient (see Sec-tion 4.3 and 5.1). If the gradient is real and is caused bygalaxy rotation, the mass estimator based on velocity dis-persion alone could be underestimating the dynamical massand therefore the mass to light ratio. To gauge the possibleimpact of this we make a simple modification of the Walkeret al. (2009b) estimator, adding a rotation term

M(< r1/2) =5

2G

(σ2 +

2

5V 2rot

)r1/2

Assuming that the rotation velocity can be obtained fromthe radial velocity gradient evaluated at half light radiusVrot = dV

dα cos δcenr1/2 and that we observe the rotation in

the edge-on orientation, we can obtain a half-light dynami-cal mass of M(< r1/2) = 2.60+0.98

−0.70 105 M. This measure-ment is larger by ∼ 20% than the dynamical mass calculatedwithout taking into account the rotation and would implythe mass-to-light ratio of ∼ 80. This is not enough, however,to put Hydrus 1 in line with other ultra-faints on Fig. 11. Ofcourse, the impact of rotation can be increased if we assumethat the rotation plane is not viewed edge-on, in which casethe true rotation would be higher by cos i and could in prin-ciple increase the dynamical mass of Hydrus 1 to match theoverall relation in Fig. 11.

5.3 Connection to the Magellanic Clouds

As seen in Figure 1, the location of the Hydrus 1 satelliteon the sky is right in between the Large and the Small Mag-ellanic Clouds. Moreover, as discussed in Section 3.2, thesatellite orientation on the sky appears to be aligned withthe direction of the Clouds’ motion on the sky. Thus, it is notimprudent to ask: how likely is the association of Hydrus 1

MNRAS 000, 1–21 (2018)

Hydrus 1 13

−40

−20

0

20

40

60B M

S[d

eg]

020406080

100120140

D

[kpc

]

−100−50050100LMS [deg]

−400−300−200−100

0100200300400

v GSR

[km

s−1 ]

Figure 12. Density of the LMC debris in the maximum-likelihood model of Jethwa et al. (2016). Top: Probability dis-

tribution of the LMC satellites on the sky in the Magellanic coor-

dinate system as defined by Nidever et al. (2008). Black contoursshow neutral hydrogen gas in the Magellanic Stream. Large and

small black circles mark the locations of the LMC and the SMC

respectively. White star marker shows the location of Hydrus 1,while white triangles show the location of known MW satellites.

Middle: Colored density contours represent the expected proba-bility distribution of the LMC satellites in the Magellanic latitude

vs heliocentric distance space. The meaning of the symbols is the

same as in the top panel. The contours for the Magellanic streamare not shown due to the lack of distance estimates available.

Bottom: Same as above but for the galactocentric line-of-sight

velocity vs Magellanic latitude.

with the Magellanic accretion event? In fact, the Magellanichypothesis has recently been invoked to explain the natureof many of the satellites discovered in the vicinity of theClouds in the past few years (Koposov et al. 2015a,b; Drlica-Wagner et al. 2015b; Torrealba et al. 2018). Unfortunately,it is impossible to give the exact answer to this questionwithout the complete set of the satellite’s phase-space coor-dinates combined with the accurate mass estimates of theLMC and the SMC. Nonetheless, one may draw some usefulconclusions by comparing the numerical simulations of theMagellanic Clouds’ disruption to the observed properties ofthe satellite (see e.g. Nichols et al. 2011; Jethwa et al. 2016;Sales et al. 2017).

Figure 12 shows the expected distributions of the par-ticles stripped from the Large Magellanic Cloud as com-puted by Jethwa et al. (2016). The top panel gives the po-sitions on the sky in the Magellanic Stream coordinate sys-

tem (see Nidever et al. 2008). The middle panel presents thedependence of the heliocentric distance on the Magellanicstream longitude LMS. Finally, the bottom row reveals howthe galactocentric radial velocity evolves with LMS. Over-plotted on top of the distribution of the simulated particlesare the positions of the currently known MW satellites to-gether with that of Hydrus 1. Unsurprisingly, as illustratedin the top panels, on the sky, Hyi 1 sits right at the peakof the predicted debris distribution. However, the match be-tween the simulations and the observations appears to bemuch less convincing if the other two phase-space projectionsare considered. In particular, Hyi 1 is significantly closerthan the majority of the LMC debris, yet its heliocentricdistance is in marginal agreement with the maximum like-lihood model prediction. Similarly, the line-of-sight velocitymeasurement of Hyi 1 is not well reproduced by the model,but appears compatible with the broad trend of increasingVGSR as a function of LMS. It is difficult to speculate howinformative the agreement between the observed position ofHydrus 1 and the Jethwa et al. (2016) models is, given thatthe width of the LOS velocity distribution is ∼ 200 km s−1.Naively, this would imply that there is a ∼ 50% chance thata random MW halo object, not related to the MC in-fall,will possess a velocity in agreement with the Jethwa et al.(2016) predictions.

Some of the thickness of the density contours of theLMC debris in Figure 12 may be due to the maximum like-lihood model of Jethwa et al. (2016) attempting to explainthe wide spread of the positions of the DES satellites onthe sky and in the distances along the line of sight. Whilethis model generates a broad cloud of tidal debris which en-velopes most of the satellites, it appears to be incorrect in de-tail. As pointed out by Jethwa et al. (2016), the satellite dis-tribution around the LMC is rather anisotropic, with manyof the satellites forming a narrow sequence aligned with theprojection of the SMC’s orbital plane onto the sky (see Fig-ure 7 of Torrealba et al. (2018)). Therefore, given that theabove maximum likelihood model may be over-bloated, itis useful to conduct a different experiment and explore in-stead whether there exists any model of the Clouds’ accre-tion in which the observed phase-space position of Hyi 1 issignificantly more likely. Motivated by the recently reportedmeasurements giving preference to the “light” Galaxy (seee.g. Gibbons et al. 2014; Williams et al. 2017; Eadie et al.2017; Patel et al. 2018), we focus on a particular simulation,in which the MW/LMC/SMC masses are 75/5/4×1010M.For this simulation, Figure 13 shows the density of the LMCparticles in the plane spanned by the line-of-sight velocityVGSR and the distance from the LMC, RLMC, marginalizedover 150 different values of the Clouds’ proper motions. Notethat we select particles in a narrow slice of Magellanic longi-tudes corresponding to that of Hyi 1 and a range of Magel-lanic latitudes. The left panel of the Figure shows the viewof the LMC debris not too dissimilar to the ML model pre-dictions displayed in Figure 12. Here, a broad - albeit lumpy- VGSR distribution is visible, without any strong trends asa function of RLMC. However, as the middle panel of Fig-ure 13 reveals, there exists a noticeable dynamical age gradi-ent across the (VGSR, RLMC) plane. More precisely, the parti-cles stripped recently, i.e. less than 0.1 Gyr ago populate lowVGSR values across a wide range of the LMC distance. How-ever, the debris unbound from the Cloud more than 0.1 Gyr

MNRAS 000, 1–21 (2018)

14 S. E. Koposov et al.

All LMC debris

-200 -100 0 100 200 300VGSR, km/s

-10

0

10

20

30

40

50R

LMC, k

pcLMC debris age

-200 -100 0 100 200 300VGSR, km/s

-10

0

10

20

30

40

50

RLM

C, k

pc

8 7 6

LMC debris with 0.1<t (Gyr)<0.3

-200 -100 0 100 200 300VGSR, km/s

-10

0

10

20

30

40

50

RLM

C, k

pc

Figure 13. Distribution of the line-of-sight velocity VGSR of the simulated LMC debris as a function of the distance from the Cloud,

RLMC. In this particular simulation, the MW/LMC/SMC masses are fixed at 75/5/4×1010M. We select particles at the same Magellaniclongitude LMS ∼ −7 as that of Hyi 1, but in a range of latitudes BMS. The location of the Hyi 1 satellite is marked with a black star.

Left: Density of particles in the VGSR, RLMC. The black histogram shows the 1D distribution of VGSR. Middle: Mean dynamical age of

the particles as a function of VGSR and RLMC. According to the greyscale bar displaying the logarithm of the stripping time in years,the bulk of the debris was removed from the LMC less than 0.5 Gyr ago. There is a clear age gradient as a function of VGSR, with older

debris in the leading (trailing) tail possessing higher negative (positive) velocities. Right: Only debris stripped between 0.1 and 0.3 Gyr

ago are shown. Histogram shows the 1D VGSR distribution for particles with 20 < RLMC/1 kpc< 30, i.e. those at distances from theLMC similar to that of Hyi 1. Note the strong bimodality of the velocity distribution.

-200 -100 0 100 200 300VGSR, km/s

0.00

0.05

0.10

0.15

Age

, Gyr

-200 -100 0 100 200 300VGSR, km/s

0.00

0.05

0.10

0.15

0.20All debris

t > 0.1 Gyr

Figure 14. Distribution of the LMC debris in the simulation

considered in Figure 13 within 10 kpc of Hyi 1. Left: Density of

particles in the plane of Galactocentric line-of-sight velocity anddynamical age (time of stripping from the LMC). Notice two dis-

tinct clouds of debris: the one corresponding to the most recently

unbound particles, those with age < 0.1 Gyr and a range of veloc-ities, and the second group with t > 0.1 Gyr. Right: 1D velocity

distribution of all debris (debris with t > 0.1 Gyr) shown in grey(black). Dashed line marks the measured VGSR of Hyi 1.

ago form tight sequences with high negative and high pos-itive velocities, likely corresponding to the leading and thetrailing tails. Hyi 1 (shown as a black star) can be found onthe leading sequence, whose VGSR velocity decreases awayfrom the LMC. As displayed in the right panel of the Fig-ure, if only dynamically old particles are selected, the his-togram of their radial velocities becomes strongly bimodalwith Hyi 1 situated near the peak of the leading debris dis-tribution.

We confirm that the old debris, i.e. those stripped longerthan 0.1 Gyr ago have radial velocities similar to that mea-sured for Hyi 1 by selecting particles within 10 kpc of thesatellite (in 3D). The distribution of the simulated debirs

is shown in Figure 14, where two clumps with distinct ageproperties are apparent (left panel). The dynamically olddebris occupy very narrow velocity range very close to VGSR

of Hyi 1 (right panel of the Figure). It therefore appears thatin the low-mass MW, some of the satellites stripped in theearly phases of the LMC disruption can populate the lead-ing debris tail (with kinematics similar to that measured forHydrus) while staying in a close proximity to the Cloud. In3D, Hydrus 1 is only 24 kpc away from the LMC. Interest-ingly, although this is nominally larger than the recent lowbounds on the tidal radius of the Cloud (Mackey et al. 2016;Besla et al. 2016), it is nonetheless quite close to it.

The final answer as to the possibility of the associa-tion of Hyi 1 and the LMC will likely be procured usingthe data from the second data release of the Gaia satel-lite, which will provide plenty of the satellite’s memberswith measured proper motion with individual uncertaintiesup to ∼ 0.1 mas yr−1. According to the ML model predic-tion of Jethwa et al. (2016), a satellite stripped from theLMC and currently observed at the position of Hyi 1 withVGSR ∼ −95 ± 30 kms−1 would have µW = −2.85 ± 1.7mas yr−1 and µN = −3.0 ± 1.8 mas yr−1 (in the Heliocen-tric frame). On the other hand, if we select only the particlescontributing to the overdensity of old debris around the loca-tion of Hyi 1 as shown in Figure 14, we obtain µW = 2.0±0.6mas yr−1 and µN = −1.0± 0.8 mas yr−1.

5.4 Dark matter annihilation

Hydrus 1’s close proximity makes it an interesting targetfor indirect searches for photons that might be releasedas products of dark matter particle interactions—e.g., an-nihilation or decay. In recent years some of the deepestsearches and strongest constraints on annihilating dark mat-ter have come from gamma-ray observations of the Milky

MNRAS 000, 1–21 (2018)

Hydrus 1 15

Way’s dwarf spheroidal galaxies, and in particular from theultrafaint population revealed by deep sky surveys (e.g. Heet al. 2015; Geringer-Sameth et al. 2015a; Ackermann et al.2015; Geringer-Sameth et al. 2015b; Drlica-Wagner et al.2015a; Li et al. 2016; Albert et al. 2017). Here we quantifythe expected signal from Hydrus 1 and perform a search forgamma-ray emission with data from the Fermi Gamma-raySpace Telescope.

For a given particle physics model (i.e., particle mass,interaction cross section), the flux of the expected photonsignal depends on the distribution of dark matter withinthe source. For annihilation the flux is proportional to theJ-profile:

dJ(n)

dΩ=

∞∫l=0

[ρ(ln)]2 dl,

where l is the line of sight distance in direction n, ρ(ln) is thedark matter density at location ln and Ω is solid angle. Fol-lowing the procedure described in detail by Geringer-Samethet al. (2015), we use the spherical Jeans equation to estimatethe dark matter density profile, ρ(r), of Hydrus 1. We adoptthe sample of 31 stars for which our measurements of posi-tion, velocity and metallicity imply membership probability> 0.95. Assuming hydrostatic equilibrium, spherical sym-metry, constant velocity anisotropy and negligible contami-nation by unresolved binary stars, we adopt the same five-parameter dark matter halo model and priors as describedby Geringer-Sameth et al. (2015). We also follow the sameprocedure to obtain the marginalized posterior probabilitydistribution for log10[J(θ)/(GeV2cm−5)], where J(θ) is theintegral of dJ/dΩ over a solid angle corresponding to angularradius θ. In particular, since the kinematic data do not con-strain the dark matter halo beyond the orbits of the star inthe dataset, we truncate ρ(r) at the radius of the outermostmember star (see Geringer-Sameth et al. (2015, Sec. 6.4)).For Hydrus 1 the estimated (deprojected) distance to theoutermost member star is 140 pc (2.6 rhalf ), correspondingto a maximum angle θmax = 0.29, beyond which dJ/dΩ iszero. This truncation gives a conservative estimate of theamplitude predicted for an annihilation signal under a givenparticle physics model.

The resulting measurement of Hydrus 1’s J value islog10 J(θmax)/(GeV2cm−5) = 18.33+0.38

−0.34 (16th, 50th, and84th percentiles of the posterior). In Fig. 15 we compareHydrus 1 with Milky Way dwarfs in the “top tier” of J val-ues. For purposes of ranking it is important to treat thedwarfs as similarly as possible when computing J ’s. Differ-ent analyses of the same kinematic data set often yield Jvalues differing by amounts comparable to the statistical er-ror. This is due to choices in the Jeans analysis (e.g. formof the dark matter profile, prior ranges, halo truncation, theassumption of spherical symmetry, etc.). In Fig. 15 all Jvalues represented by green diamonds are computed usingthe method (Geringer-Sameth et al. 2015). We omit recentlypublished J values (e.g. those of Bonnivard et al. (2015a,b);Simon et al. (2015); Zhu et al. (2016); Hayashi et al. (2016);Genina & Fairbairn (2016); Walker et al. (2016); Bonnivardet al. (2016); Klop et al. (2017); Pace & Strigari (2018)) sincethe comparison with Hydrus 1 will not be straightforward.

There have been several recent attempts to simplify the

Hyi117

18

19

20

log

10J

(0.5 )/(G

eV2cm−

5)

UMaIISeg1

ComaDraco

UMiScu

TucIIHorI

RetII

Hyi1 (this work) and others (GSKW15)

J ∝ σ4LOS/D

2rhalf (Pace & Strigari 2018)

J ∝ D−2 (Albert et. al. 2017)

Figure 15. The astrophysical J factor for dark matter annihila-

tion of Hydrus 1 compared with those of other Milky Way dwarf

spheroidals. Green diamonds show J of several dwarfs deter-mined with the procedure given in Geringer-Sameth et al. (2015)

(GSKW15). Blue circles and gray triangles are estimates of J

based on simple scaling formulas. They are fit to the GSKW15J ’s but their values for Hydrus 1 can be considered predictions.

estimation of J values using functions of photometrically-determined properties (distance, half-light radius) and,sometimes, the total line-of-sight velocity dispersion. Theseformulas usually have free parameters that are fit to exist-ing compilations of J values based on the Jeans equation.Hydrus 1 presents the first chance to test them. In Fig. 15we show two predictions for the J value for Hydrus 1. Bluecircles are based on the formula J ∝ σ4

losD−2r−1

half (Pace &Strigari 2018), whose form can be motivated by simple an-alytic considerations (e.g. Evans et al. 2016). In addition tocalibrating to their own Jeans analysis of the dwarfs, Pace& Strigari (2018) recalibrate the formula to the Geringer-Sameth et al. (2015) data set (green diamonds), finding thatit fits dwarf J ’s with residual error of 0.08 dex. Adoptingthe latter calibration, we see that the scaling relation ap-pears to fit the measured value for Hydrus 1 fairly well (er-ror bars come from adding in quadrature the uncertainty invelocity dispersion, distance, half-light radius and a 0.08 dexintrinsic scatter). The gray triangles in Fig. 15 correspondto a purely photometric determination of J , developed byDrlica-Wagner et al. (2015a); Albert et al. (2017), where Jis a function of distance alone, scaling as J ∝ D−2. As withthe blue circles, this relation was calibrated to the resultsof Geringer-Sameth et al. (2015). The distance-based J es-timate severely overpredicts Hydrus 1’s J value. While thefailure of such a simple estimator is perhaps not surprising,the lesson is that spectroscopic observations of dwarf galax-ies are necessary to determine their dark matter content. Ifthe distance-based scaling is used to predict dwarf J val-ues it must be accompanied by a large error bar (Albertet al. 2017). Here, an intrinsic uncertainty > 0.8 dex wouldbe needed to bring the estimate to < 1σ tension with theobservationally determined J of Hydrus 1.

The relative J values among different dwarfs are an es-sential ingredient in joint searches for an annihilation signal

MNRAS 000, 1–21 (2018)

16 S. E. Koposov et al.

0.00 0.05 0.10 0.15 0.20 0.25 0.30

θ [deg]

0.0

0.2

0.4

0.6

0.8

1.0J

conta

inm

ent

fract

ion

CTA (1 TeV)

Fermi-LAT (1 GeV)

Figure 16. Constraints on the spatial extent of Hydrus 1’s dark

matter annihilation profile. For each angle θ away from the cen-

ter of the dwarf the J containment fraction gives the fraction ofannihilation flux that comes from the region within θ. The black

dotted line is the median estimate, while the gray band shows the

16th to 84th percentiles of the posterior. The gamma-ray PSFsof the CTA (at 1 TeV) and Fermi-LAT (at 1 GeV) are shown for

comparison. CTA will detect Hydrus 1 as an extended source.

from all dwarfs at once (e.g. Geringer-Sameth & Koushi-appas 2011; Ackermann et al. 2011; Mazziotta et al. 2012).Pointed instruments (e.g. imaging atmospheric Cherenkovtelescopes) must incorporate the J rankings in designing ob-serving strategies. Ratios of J ’s also provide a crucial test ofa dark matter origin for any signal detected from a dwarf,since fluxes must be proportional to J (Strigari et al. 2007;Geringer-Sameth et al. 2015; Geringer-Sameth et al. 2015b;Albert et al. 2017). Although Hydrus 1 is not among thetop tier of dwarfs in terms of expected annihilation flux itoutranks most classical and ultrafaint dwarfs that are com-monly used to constrain dark matter particle physics.

Hydrus 1 is close enough to earth that its dark mat-ter annihilation may have a detectable spatial extent. Adetection in gamma-rays would be a direct observation ofthe shape of Hydrus 1’s dark matter density profile. Follow-ing Geringer-Sameth et al. (2015), Fig. 16 illustrates Hy-drus 1’s spatial extent. For each angle θ, the shaded regionshows a 68% credible interval on the fraction of annihilationflux coming from angles less than θ. Approximate PSFs forFermi-LAT and CTA are shown for comparison8. If CTAis able to detect Hydrus 1 in the future it will almost cer-tainly be seen as an extended source. The dark matter halotruncation radius we adopt is much smaller than the scaleof the Fermi-LAT PSF. If Hydrus 1’s halo extends well be-yond the outermost spectroscopically-observed member starit may appear extended to Fermi. However, this question isbeyond the scope of our conservative analysis.

Finally, we perform a search for gamma-ray emissionfrom Hydrus 1 using data from the Fermi Large Area Tele-

8 They are modeled as 2-d Gaussian distributions with 68% con-

tainment fractions of 0.05 for CTA at 1 TeV and 0.8 for Fermi-LAT at 1 GeV.

scope (Atwood et al. 2009). Hydrus is located 0.45 in pro-jection from a gamma-ray source 3FGL J0224.1-7941 in theThird Fermi Source Catalog (3FGL) (Acero et al. 2015),which is associated with a probable blazar. This precludesperforming the search using optimized event-weighting (e.g.Geringer-Sameth et al. 2015a,b) as that method does nottake into account contamination from the bright nearbysource. We therefore adopt the commonly used approachbased on maximum likelihood modeling of the region sur-rounding Hydrus 1. A best-fitting model is found both withand without a trial source at the location of Hydrus 1. Ifthe likelihood is sufficiently improved by the addition of thetrial source it is interpreted as evidence for emission fromthe direction of Hydrus 1.

Gamma-ray events from a 10× 10 region centered onHydrus 1 are binned into 0.05 spatial pixels and 25 loga-rithmic energy bins from 1 GeV to 316 GeV (10 bins perdecade). The spatial binning is substantially smaller thanthe Fermi PSF at all energies above 1 GeV and the analysiswill be as sensitive as if we had used an unbinned likeli-hood function. Restricting the energy range to be greaterthan 1 GeV avoids events with poor direction reconstruc-tion and does not negatively impact the power of the searchfor dark matter for particle masses greater than a fewGeV (Geringer-Sameth et al. 2015a, Sec. VI D). Our dataset comprises 9.4 years of observation9. The likelihood func-tion is the product of an independent poisson probabilityfor each bin. The predicted number of events in a bin is thesum of isotropic10 and diffuse11 background components12

along with the four 3FGL sources within the 10 × 10 re-gion. These sources have all been associated with extragalac-tic objects by the Fermi Collaboration in the 3FGL andwe adopt the precise locations of their counterparts ratherthan their coordinates in the 3FGL. Each source is mod-eled as a point source with a “log-parabola” energy spec-trum, whose flux takes the form dF/dE ∝ E−(α+β logE).The null model has 14 free parameters: normalizations ofthe two background components and the normalization, α,and β of each source. The trial point source located atthe position of Hydrus 1 has an energy spectrum given bydF/dE ∝ E−(α+β logE) exp(−E/Ec), with 4 additional freeparameters (normalization, α, β, and Ec). This flexible en-ergy spectrum is sufficient to describe dark matter annihila-tion into Standard Model leptons, quarks, and gauge bosons.

The test statistic is λobs = 2 log(L1/L0), where L0 isthe value of the likelihood maximized over the 14 free pa-

9 We use version v10r0p5 of the Fermi Science Tools. Front andback P8R2_SOURCE_V6 events from weeks 9 to 500 were extractedwith gtselect (zmax=90), and gtmktime (filter="(DATA_QUAL>0)

&& (LAT_CONFIG==1)" and roicut=no). We ran gtltcube with

zmax=90 and gtexpcube2 with binsz=0.1 to create an exposuremap, gtpsf to compute the PSF at the location of Hydrus 1,

and assume that the PSF is constant across region surroundingHydrus 1.10 iso_P8R2_SOURCE_V6_v06.txt11 gll_iem_v06.fits12 At the edges of each energy bin the background templateswere multiplied by the exposure map, convolved with the PSF

at a resolution of 0.05 over a 25 × 25 region, and integrated

over the energy bin (assuming log flux ∝ log energy) to obtainthe model prediction.

MNRAS 000, 1–21 (2018)

Hydrus 1 17

rameters of the null model without Hydrus 1 and L1 is themaximum likelihood value when all 18 parameters are freeto vary. The detection significance (p value) is the probabil-ity that, if the null model were true, λ would be measuredto be larger than λobs. After separately maximizing over the14 and 18 parameter model we measure λobs = 1.73. Cal-ibrating the distribution of λ using the region surroundingHydrus 1, we find the probability of obtaining a larger λunder the null model is & 21% and conclude that the Fermidata do not show any evidence for gamma-ray emission fromHydrus 1. This is consistent with gamma-ray observations ofdwarfs with larger J values (e.g. Albert et al. 2017), includ-ing both non-detections as well as a potential gamma-raysignal from Reticulum II if the latter is due to dark mat-ter annihilation (Geringer-Sameth et al. 2015b; Hooper &Linden 2015) (see also Drlica-Wagner et al. (2015a); Albertet al. (2017)). We also search for new sources in the vicinityof Hydrus 1 by adding a trial source at each location withinthe 10 × 10 region. This search reveals two probable newgamma-ray sources but they are several degrees away fromHydrus 1.

5.5 Chemistry

In Section 4.4 we described that we were able to get robustabundances of more than one chemical element, in particu-lar iron, magnesium and carbon for many Hydrus members.First we look at the abundances of alpha elements vs ironelements in the satellite. The alpha element abundances areparticularly useful for constraining the length of star forma-tion history, due to the fact that Type Ia and Type II Su-pernovae have vastly different yields in alpha-elements andthat Type Ia supernovae occur with a delay of ∼ 100 Myrafter the beginning of star formation and the first Type IIsupernovae explosions (Woosley & Weaver 1995; Maoz et al.2014). Figure 17 shows the [Mg/Fe] vs [Fe/H] abundancedistribution for Hydrus 1 member stars, together with thedistribution in the MW halo and other UFDs. We note that,on average, the [Mg/Fe] in Hydrus’s stars is larger than zero<[Mg/Fe]>∼ 0.2, which is indicative that the star-formationepisode that led to the formation of Hydrus 1 was not veryextended in time. However, the data also show that [Mg/Fe]decreases significantly from [Mg/Fe]∼ 0.6 at [Fe/H] ∼ −3to almost solar at higher metallicity. This feature is asso-ciated with self-enrichment of gas by Type Ia supernovae(Tinsley 1979) and implies that the star formation proceededfor at least ∼ 100 Myr so that the α abundances were di-luted by Type Ia yields. The lack of higher metallicity starswith solar alpha-to-iron abundances, however, means thatthe star formation did not proceed longer than 100 Myr-1 Gyr. A similar decline in alpha abundances is very pro-nounced in the MW halo and MW thick disk, as well asin some classical dwarfs (Tolstoy et al. 2009) and more lu-minous UFDs (Vargas et al. 2013), where it tends to occurat higher metallicities. In Hydrus 1 this knee occurs at ametallicity of −2.6.

On the bottom panel of Figure 17 we also show themetallicity distribution function (MDF) from spectral syn-thesis modeling of Hyi 1 stars. The distribution seems tobe asymmetric with a longer tail towards high metallicities,thereby challenging our assumption of Gaussian metallic-ity distribution for the purposes of our chemo-dynamical

[Fe/H]

−0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

[Mg/

Fe]

MWdSph,Vargas+2013Hydrus 1

−3.5 −3.0 −2.5 −2.0 −1.5 −1.0

[Fe/H]

0.0

2.5

5.0

Nst

ars/

bin

Figure 17. Chemical abundances of Hydrus 1 member stars. Top

panel: [Mg/Fe] vs [Fe/H] for Hyi 1 stars are shown by red pointswith ellipses showing the 1 sigma covariance. Milky Way stars

from the SAGA database (Suda et al. 2008) are shown by small

light grey markers while stars from several faint dwarf spheroidals(Vargas et al. 2013) are shown by large dark grey circles. One

Hydrus 1 star (star 26) is outside the plot, with [Fe/H]∼ −2.15

and [Mg/Fe]∼ −0.9 Bottom panel: Histogram of iron abundancesof Hydrus 1 stars. The dashed lines show the mean and stan-

dard deviation of the Gaussian abundance distribution that was

part of our chemo-dynamical model. We note that the metallicitydistribution seems to be non-Gaussian, with a more pronounced

metal-rich tail.

modelling. However, because of small number statistics andsignificant uncertainties in individual metallicity measure-ments, it remains to be seen whether the MDF asymme-try is real. We also remark that most dwarf galaxies showMDFs with a more pronounced metal-poor tail; however,some, such as Ursa Minor (Kirby et al. 2011), show a metalrich tail, which could be explained by chemical evolutionmodels with lower galactic wind efficiency.

Two likely Hyi 1 member stars with higher metallici-ties shown in Fig. 17 have high [Mg/Fe] ratio. The cause ofthis is unclear. It is possible, although unlikely, that thoseare MW halo contaminants rather than Hydrus 1 stars. Wealso remark on one star (star 26) that is outside the plot-ting range on Figure 17 with unusually low [Mg/Fe]∼ −0.9.Similar abundances were previously observed in some chem-ically peculiar stars in other dwarfs, such as Carina (Venn etal. 2012), where the unusual abundance patters were poten-tially associated with inhomogeneous gas mixing and starformation from SN Ia enriched gas, leading to an overabun-dance of iron with respect to other elements.

Upper limits on [C/H] are available for 21 Hydrus 1members. These limits are not informative. We can onlyconfirm that one of the more metal-rich stars, Star 77 with[Fe/H] = −1.60 ± 0.10, is not a carbon-enhanced metal-poor star: [C/Fe] < 0.54. For all other members with up-per limits on [C/H], the limits are higher than both the[C/Fe] > +0.7 and [C/Fe] > +1.0 definitions that are com-monly adopted for carbon-enhanced metal-poor stars (Aoki

MNRAS 000, 1–21 (2018)

18 S. E. Koposov et al.

et al. 2007; Beers & Christlieb 2005). From these limits wecannot summarise how many Hydrus 1 members meet thedefinition for carbon enhancement.

With [C/Fe] ∼ +3 and [Fe/H] = −3.18, Star 75 clearlymeets the definition of a carbon-enhanced metal-poor star.Indeed, the estimated [C/H] ratio is near the Solar value,which is remarkable for an extremely metal-poor star in anultra-faint dwarf galaxy. A search of the SAGA database(Suda et al. 2008) and the JINAbase (Abohalima & Frebel2018) reveals that Star 75 of Hydrus 1 is the most carbon-enhanced metal-poor star known in a dwarf galaxy. Thenext most carbon-enhanced stars known in dwarf galaxiesare Star 11 in Bootes I (Lai et al. 2011), and Star 7 in Segue1 (Frebel et al. 2014), both which have [C/Fe] ≈ +2.3. Sim-ilarly, in a study of 67 carbon-enhanced metal-poor stars inthe Milky Way halo, only one was found to have [C/Fe] > 3(Abate et al. 2015).

Strong carbon enhancement in metal-poor stars resultsin depressed flux levels at bluer wavelengths (making g rel-atively fainter than r), causing a reddening in the g − rcolour. This is consistent with what we observe in the sec-ond panel of Figure 5. Star 75, as marked with a cross-hair,is some ∼0.1 magnitudes redder than most other members.This effect has been discussed before (Aoki et al. 2002). Sincecandidate members in ultra-faint dwarf galaxies are usuallyselected for follow-up spectroscopy based on their proximityto an isochrone (among other factors), it is likely that anysimilarly carbon-enhanced metal-poor stars in other ultra-faint dwarf galaxies have been overlooked. This acts to fur-ther complicate any estimates of the carbon-enhanced metal-poor star fraction among these near pristine galaxies (Frebel& Norris 2015).

6 CONCLUSIONS

In this paper we presented the discovery and detailed anal-ysis of a new ultra-faint Milky Way satellite, called Hydrus1. Below we summarize the key findings about the system.

• The satellite is located in between the Large and theSmall Magellanic Clouds on the sky at the heliocentric dis-tance of 28 kpc and 24 kpc from the LMC. The luminosity ofthe satellite is MV ∼ −4.7 and the color-magnitude diagramindicates an old, metal-poor stellar population with severalBHB stars and two RR-Lyrae. The object is mildly ellipti-cal with an axis ratio of 0.8, and is approximately alignedwith the right ascension axis and proper motion of LMC.The circularized half-light radius of Hydrus 1 is 50 pc, andit has a surface brightness of ∼ 27.5 mag/sq. arcsec, whichis noticeably brighter than many typical ultra-faint dwarfspheroidals. These properties place the object in an am-biguous region between extended globular clusters and dwarfgalaxies in the plane of size and luminosity.• High-resolution spectroscopy revealed ∼ 30 confirmed

members of Hydrus 1 with mean VGSR of −94 km s−1 and awell resolved stellar velocity dispersion of 2.7 km s−1. Theimplied mass-to-light ratio within the half-light radius islog10 M/L = 1.82± 0.16, thus indicating that the system isdark matter dominated. Nevertheless, the mass-to-light ra-tio is uncharacteristically low in comparison to other dwarfsof similar luminosity. We speculate that it is possible thatmany other Galactic faint satellites with ambiguous sizes

between extended clusters and dwarf galaxies also have in-termediate values of mass-to-light ratios. These systems mayeither be globular clusters with a small amount dark matteror dwarf galaxies that lost some dark matter due to tides.Our analysis of the stellar kinematics of Hydrus 1 also re-veals a tentative (at 2.5 sigma significance level) line-of-sightvelocity gradient, which is compatible with a 3 km s−1 rota-tion of the satellite. If real, this could signify the detectionof a first rotating ultra-faint dwarf galaxy.• Based on the spatial position and the observed line-of-

sight velocity, we find it highly plausible that Hydrus 1 ispart of the Magellanic family of satellites. The dwarf maycurrently be a part of the leading debris of the LMC, havingbeen stripped from the LMC between 0.1 Gyr and 0.5 Gyrago. A connection between Hydrus 1 and the LMC shouldbe confirmed or refuted by the incoming Gaia DR2 data.• Due to its relative proximity to the Sun, Hydrus 1 is a

very promising target for dark-matter annihilation searches,with a J-factor of log J = 18.33 ± 0.36. While it is not thehighest J-factor among all the MW dwarfs, it is competitiveand has a small uncertainty, thanks to the high quality veloc-ity dispersion measurement. Our analysis of the Fermi dataat the location of Hydrus for the dark matter annihilationreveals no significant signal.• Chemically, Hydrus 1 is rather metal poor, with mean

[Fe/H] ∼ −2.5 and significant scatter in abundances of0.4 dex. It has an enhanced alpha-abundance <[Mg/Fe]>∼0.2 and shows evidence of self-enrichment by SN Ia attimescales of 100-1000 Myr demonstrated by the decreas-ing [Mg/Fe] with increasing [Fe/H]. Analysis of 30 Hy-drus 1 member stars revealed one carbon-enhanced ex-tremely metal-poor star with metallicity [Fe/H]<−3 and[C/Fe] ∼ 3. This level of carbon enrichment makes it onethe most carbon-enhanced stars known in both the MW haloand the dwarf galaxies.

ACKNOWLEDGEMENTS

A. R. C. acknowledges support from the Australian Re-search Council (ARC) through Discovery Project grantDP160100637. D.M. is supported by an ARC Future Fel-lowship (FT160100206). D.M. and G.D.C. acknowledge sup-port from ARC Discovery Project DP150103294. M.M. ac-knowledges support from NSF grant AST1312997. M.G.W.acknowledges support from NSF grant AST1412999. E.O.acknowledges support from NSF grant AST1313006. Theresearch leading to these results has received funding fromthe European Research Council under the European Union’sSeventh Framework Programme (FP/2007-2013) / ERCGrant Agreement n. 308024.

This project used data obtained with the Dark En-ergy Camera (DECam), which was constructed by the DarkEnergy Survey (DES) collaboration. Funding for the DESProjects has been provided by the U.S. Department of En-ergy, the U.S. National Science Foundation, the Ministry ofScience and Education of Spain, the Science and Technol-ogy Facilities Council of the United Kingdom, the HigherEducation Funding Council for England, the National Cen-ter for Supercomputing Applications at the University ofIllinois at Urbana-Champaign, the Kavli Institute of Cos-mological Physics at the University of Chicago, the Center

MNRAS 000, 1–21 (2018)

Hydrus 1 19

for Cosmology and Astro-Particle Physics at the Ohio StateUniversity, the Mitchell Institute for Fundamental Physicsand Astronomy at Texas A&M University, Financiadorade Estudos e Projetos, Fundacao Carlos Chagas Filho deAmparo a Pesquisa do Estado do Rio de Janeiro, Con-selho Nacional de Desenvolvimento Cientıfico e Tecnologicoand the Ministerio da Ciencia, Tecnologia e Inovacao, theDeutsche Forschungsgemeinschaft, and the CollaboratingInstitutions in the Dark Energy Survey. The CollaboratingInstitutions are Argonne National Laboratory, the Univer-sity of California at Santa Cruz, the University of Cam-bridge, Centro de Investigaciones Energeticas, Medioambi-entales y Tecnologicas-Madrid, the University of Chicago,University College London, the DES-Brazil Consortium,the University of Edinburgh, the Eidgenossische TechnischeHochschule (ETH) Zurich, Fermi National Accelerator Lab-oratory, the University of Illinois at Urbana-Champaign, theInstitut de Ciencies de l’Espai (IEEC/CSIC), the Institut deFısica d’Altes Energies, Lawrence Berkeley National Labo-ratory, the Ludwig-Maximilians Universitat Munchen andthe associated Excellence Cluster Universe, the Universityof Michigan, the National Optical Astronomy Observatory,the University of Nottingham, the Ohio State University, theOzDES Membership Consortium the University of Pennsyl-vania, the University of Portsmouth, SLAC National Ac-celerator Laboratory, Stanford University, the University ofSussex, and Texas A&M University.

This work has made use of data from the EuropeanSpace Agency (ESA) mission Gaia (https://www.cosmos.esa.int/gaia), processed by the Gaia Data Processing andAnalysis Consortium (DPAC, https://www.cosmos.esa.

int/web/gaia/dpac/consortium). Funding for the DPAChas been provided by national institutions, in particular theinstitutions participating in the Gaia Multilateral Agree-ment.

Based on observations at Cerro Tololo Inter-AmericanObservatory, National Optical Astronomy Observatory(NOAO Prop. 2016A-0618 and 2017B-0906; PI: D. Mackey),which is operated by the Association of Universities for Re-search in Astronomy (AURA) under a cooperative agree-ment with the National Science Foundation.

REFERENCES

Abate, C., Pols, O. R., Izzard, R. G., & Karakas, A. I. 2015, A&A,581, A22

Abbott, T. M. C., Abdalla, F. B., Allam, S., et al. 2018,arXiv:1801.03181

Abohalima & Frebel 2018, submitted (arXiv:1711.04410)

Acero F., et al., 2015, ApJS, 218, 23

Ackermann M., et al., 2011, Physical Review Letters, 107, 241302

Ackermann M., et al., 2015, Physical Review Letters, 115, 231301

Aden, D., Wilkinson, M. I., Read, J. I., et al. 2009, ApJ, 706,

L150

Albert A., et al., 2017, ApJ, 834, 110

Allende Prieto, C., Sivarani, T., Beers, T. C., et al. 2008, AJ, 136,

2070

Aoki, W., Norris, J. E., Ryan, S. G., Beers, T. C., & Ando, H.2002, ApJ, 567, 1166

Aoki, W., Beers, T. C., Christlieb, N., et al. 2007, ApJ, 655, 492

Atwood W. B., et al., 2009, ApJ, 697, 1071

Bechtol, K., Drlica-Wagner, A., Balbinot, E., et al. 2015, ApJ,807, 50

Beers, T. C., & Christlieb, N. 2005, ARA&A, 43, 531

Belokurov, V., Zucker, D. B., Evans, N. W., et al. 2007, ApJ, 654,

897

Belokurov, V., Irwin, M. J., Koposov, S. E., et al. 2014, MNRAS,441, 2124

Belokurov, V., & Koposov, S. E. 2016, MNRAS, 456, 602

Belokurov, V., Erkal, D., Deason, A. J., et al. 2017, MNRAS, 466,

4711

Belokurov, V., Deason, A. J., Koposov, S. E., et al. 2018, MNRAS,

Benson, A. J., Frenk, C. S., Lacey, C. G., Baugh, C. M., & Cole,

S. 2002, MNRAS, 333, 177

Bertin, E. 2010, Astrophysics Source Code Library, ascl:1010.068

Bertin, E. 2011, Astronomical Data Analysis Software and Sys-

tems XX, 442, 435

Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393

Besla, G., Martınez-Delgado, D., van der Marel, R. P., et al. 2016,ApJ, 825, 20

Bonifacio, P., Caffau, E., Zaggia, S., et al. 2015, A&A, 579, L6

Bonnivard V., et al., 2015a, MNRAS, 453, 849

Bonnivard V., et al., 2015b, ApJ, 808, L36

Bonnivard V., Maurin D., Walker M. G., 2016, MNRAS, 462, 223

Bressan, A., Marigo, P., Girardi, L., et al. 2012, MNRAS, 427,

127

Brodie, J. P., Romanowsky, A. J., Strader, J., & Forbes, D. A.

2011, AJ, 142, 199

Brooke, J. S. A., Bernath, P. F., Western, C. M., et al. 2013,68th International Symposium on Molecular Spectroscopy,

#ERD02

Bullock, J. S., & Boylan-Kolchin, M. 2017, ARA&A, 55, 343

Caldwell, N., Walker, M. G., Mateo, M., et al. 2017, ApJ, 839, 20

Carpenter, B, Gelman, A., Hoffman, M., et al. 2017, Journal ofStatistical Software, 76(1)

Casey, A. R. 2014, Ph.D. Thesis,

Castelli, F., & Kurucz, R. L. 2004, arXiv:astro-ph/0405087

Catelan, M., Pritzl, B. J., & Smith, H. A. 2004, ApJS, 154, 633

Chabrier, G. 2003, PASP, 115, 763

Choi, J., Dotter, A., Conroy, C., et al. 2016, ApJ, 823, 102

Collins, M. L. M., Tollerud, E. J., S and, D. J., et al. 2017, MN-RAS, 467, 573

The Dark Energy Survey Collaboration 2005, arXiv:astro-

ph/0510346

Deason, A. J., Belokurov, V., & Evans, N. W. 2011, MNRAS,

416, 2903

Deason, A. J., Wetzel, A. R., Garrison-Kimmel, S., & Belokurov,

V. 2015, MNRAS, 453, 3568

Desai, S., Armstrong, R., Mohr, J. J., et al. 2012, ApJ, 757, 83

Drlica-Wagner A., et al., 2015, ApJ, 809, L4

Drlica-Wagner, A., Bechtol, K., Rykoff, E. S., et al. 2015, ApJ,813, 109

Drlica-Wagner, A., Bechtol, K., Allam, S., et al. 2016, ApJ, 833,L5

Dotter, A. 2016, ApJS, 222, 8

Eadie, G. M., Springford, A., & Harris, W. E. 2017, ApJ, 835,167

Evans N. W., Sanders J. L., Geringer-Sameth A., 2016, Phys.Rev. D, 93, 103512

Fermani, F., & Schonrich, R. 2013, MNRAS, 430, 1294

Fiorentino, G., Bono, G., Monelli, M., et al. 2015, ApJ, 798, L12

Flaugher, B., Diehl, H. T., Honscheid, K., et al. 2015, AJ, 150,

150

Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J.

2013, PASP, 125, 306

Frebel, A., Simon, J. D., & Kirby, E. N. 2014, ApJ, 786, 74

Frebel, A., & Norris, J. E. 2015, ARA&A, 53, 631

Gaia Collaboration, Prusti, T., de Bruijne, J. H. J., et al. 2016,A&A, 595, A1

Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. 2016,

A&A, 595, A2

MNRAS 000, 1–21 (2018)

20 S. E. Koposov et al.

Gelman, A., & Rubin, D. (1992). Inference from Iterative Sim-ulation Using Multiple Sequences. Statistical Science, 7(4),

457-472.

Genina A., Fairbairn M., 2016, MNRAS, 463, 3630

Geringer-Sameth A., Koushiappas S. M., 2011, Phys. Rev. Lett.,

107, 241303

Geringer-Sameth A., Koushiappas S. M., Walker M. G., 2015a,Phys. Rev. D, 91, 083535

Geringer-Sameth A., Walker M. G., Koushiappas S. M., Koposov

S. E., Belokurov V., Torrealba G., Evans N. W., 2015b, Phys.

Rev. Lett., 115, 081101

Geringer-Sameth, A., Koushiappas, S. M., & Walker, M. 2015,ApJ, 801, 74

Gibbons, S. L. J., Belokurov, V., & Evans, N. W. 2014, MNRAS,

445, 3788

Gilmore, G., Wilkinson, M. I., Wyse, R. F. G., et al. 2007, ApJ,

663, 948

Goodman, J., & Weare, J. 2010, Communications in AppliedMathematics and Computational Science, Vol. 5, No. 1, p. 65-

80, 2010, 5, 65

Gorski, K. M., Hivon, E., Banday, A. J., et al. 2005, ApJ, 622,

759

Jethwa, P., Erkal, D., & Belokurov, V. 2016, MNRAS, 461, 2212

Jethwa, P., Erkal, D., & Belokurov, V. 2018, MNRAS, 473, 2060

Jethwa, P., Torrealba, G., Navarrete, C., et al. 2017,

arXiv:1711.09103

Hayashi K., Ichikawa K., Matsumoto S., Ibe M., Ishigaki M. N.,Sugai H., 2016, MNRAS, 461, 2914

He C., Bechtol K., Hearin A. P., Hooper D., 2015, Phys. Rev. D,

91, 063515

Heiter, U., Lind, K., Asplund, M., et al. 2015, Phys. Scr., 90,

054010

Henden, A. A., Levine, S. E., Terrell, D., Smith, T. C., & Welch,D. 2012, Journal of the American Association of Variable Star

Observers (JAAVSO), 40, 430

Hoffman, M. D., & Gelman, A. 2011, arXiv:1111.4246

Hooper D., Linden T., 2015, J. Cosmology Astropart. Phys., 9,

16

Kaiser, N., Aussel, H., Burke, B. E., et al. 2002, Proc. SPIE, 4836,154

Kaplinghat, M., & Strigari, L. E. 2008, ApJ, 682, L93

Kazantzidis, S., Lokas, E. L., Callegari, S., Mayer, L., & Mous-

takas, L. A. 2011, ApJ, 726, 98

Kirby, E. N., Guhathakurta, P., & Sneden, C. 2008, ApJ, 682,

1217-1233

Kirby, E. N., Cohen, J. G., Guhathakurta, P., et al. 2013, ApJ,779, 102

Klypin, A., Kravtsov, A. V., Valenzuela, O., & Prada, F. 1999,

ApJ, 522, 82

Kim, D., Jerjen, H., Milone, A. P., Mackey, D., & Da Costa, G. S.

2015, ApJ, 803, 63

Kim, D., & Jerjen, H. 2015, ApJ, 808, L39

Kirby, E. N., Lanfranchi, G. A., Simon, J. D., Cohen, J. G., &Guhathakurta, P. 2011, ApJ, 727, 78

Kirby, E. N., Simon, J. D., & Cohen, J. G. 2015, ApJ, 810, 56

Klop N., Zandanel F., Hayashi K., Ando S., 2017, Phys. Rev. D,

95, 123012

Koposov, S., & Bartunov, O. 2006, Astronomical Data Analysis

Software and Systems XV, 351, 735

Koposov, S., Belokurov, V., Evans, N. W., et al. 2008, ApJ, 686,279-291

Koposov, S. E., Yoo, J., Rix, H.-W., et al. 2009, ApJ, 696, 2179

Koposov, S. E., Gilmore, G., Walker, M. G., et al. 2011, ApJ, 736,146

Koposov, S. E., Belokurov, V., Torrealba, G., & Evans, N. W.

2015, ApJ, 805, 130

Koposov, S. E., Casey, A. R., Belokurov, V., et al. 2015, ApJ,811, 62

Kupper, A. H. W., Kroupa, P., Baumgardt, H., & Heggie, D. C.2010, MNRAS, 401, 105

Kupper, A. H. W., Johnston, K. V., Mieske, S., Collins, M. L. M.,

& Tollerud, E. J. 2017, ApJ, 834, 112

Laevens, B. P. M., Martin, N. F., Sesar, B., et al. 2014, ApJ, 786,

L3

Laevens, B. P. M., Martin, N. F., Bernard, E. J., et al. 2015, ApJ,813, 44

Lai, D. K., Lee, Y. S., Bolte, M., et al. 2011, ApJ, 738, 51

Lee, Y. S., Beers, T. C., Sivarani, T., et al. 2008, AJ, 136, 2022

Lee, Y. S., Beers, T. C., Sivarani, T., et al. 2008, AJ, 136, 2050

Li S., et al., 2016, Phys. Rev. D, 93, 043518

Li, T. S., Simon, J. D., Pace, A. B., et al. 2018, arXiv:1802.06810

Mackey, A. D., Koposov, S. E., Erkal, D., et al. 2016, MNRAS,459, 239

Mackey, A. D., Koposov, S. E., Da Costa, G. S., et al. 2017,MNRAS, 472, 2975

Maoz, D., Mannucci, F., & Nelemans, G. 2014, ARA&A, 52, 107

Marigo, P., Girardi, L., Bressan, A., et al. 2017, ApJ, 835, 77

Martin, N. F., de Jong, J. T. A., & Rix, H.-W. 2008, ApJ, 684,

1075-1092

Martin, N. F., & Jin, S. 2010, ApJ, 721, 1333

Martin, N. F., Nidever, D. L., Besla, G., et al. 2015, ApJ, 804, L5

Mayer, L., Governato, F., Colpi, M., et al. 2001, ApJ, 559, 754

Mazziotta M. N., Loparco F., de Palma F., Giglietto N., 2012,

Astroparticle Physics, 37, 26

McConnachie, A. W. 2012, AJ, 144, 4

McMillan, P. J. 2017, MNRAS, 465, 76

Neal, R. M. 2012, arXiv:1206.1901

Nichols, M., Colless, J., Colless, M., & Bland-Hawthorn, J. 2011,

ApJ, 742, 110

Nidever, D. L., Majewski, S. R., & Butler Burton, W. 2008, ApJ,679, 432-459

Pace A. B., Strigari L. E., 2018, arxiv:1802.06811

Patel, E., Besla, G., Mandel, K., & Sohn, S. T. 2018,

arXiv:1803.01878

Paxton, B., Bildsten, L., Dotter, A., et al. 2011, ApJS, 192, 3

Penarrubia, J., Navarro, J. F., & McConnachie, A. W. 2008, ApJ,

673, 226-240

Pietrukowicz, P., Koz lowski, S., Skowron, J., et al. 2015, ApJ,

811, 113

Protassov R., van Dyk D. A., Connors A., Kashyap V. L., Siemigi-nowska A., 2002, ApJ, 571, 545

Ram, R. S., Brooke, J. S. A., Bernath, P. F., Sneden, C., & Lu-

catello, S. 2014, ApJS, 211, 5

Ricotti, M., Parry, O. H., & Gnedin, N. Y. 2016, ApJ, 831, 204

Sales, L. V., Navarro, J. F., Kallivayalil, N., & Frenk, C. S. 2017,

MNRAS, 465, 1879

Schlafly, E. F., & Finkbeiner, D. P. 2011, ApJ, 737, 103

Schlafly, E. F., Green, G. M., Lang, D., et al. 2018, ApJS, 234,39

Schlegel, D. J., Finkbeiner, D. P., & Davis, M. 1998, ApJ, 500,525

Schonrich, R., Binney, J., & Dehnen, W. 2010, MNRAS, 403, 1829

Simon, J. D., & Geha, M. 2007, ApJ, 670, 313

Simon J. D., et al., 2015, ApJ, 808, 95

Shanks, T., Metcalfe, N., Chehade, B., et al. 2015, MNRAS, 451,4238

Sneden, C. A. 1973, Ph.D. Thesis,

Skrutskie, M. F., Cutri, R. M., Stiening, R., et al. 2006, AJ, 131,

1163

Smolec, R. 2005, Acta Astron., 55, 59

Strigari L. E., Koushiappas S. M., Bullock J. S., Kaplinghat M.,

2007, Phys. Rev. D, 75, 083526

Suda, T., Katsuta, Y., Yamada, S., et al. 2008, PASJ, 60, 1159

Tinsley, B. M. 1979, ApJ, 229, 1046

Tollerud, E. J., Bullock, J. S., Strigari, L. E., & Willman, B. 2008,ApJ, 688, 277-289

MNRAS 000, 1–21 (2018)

Hydrus 1 21

Tolstoy, E., Hill, V., & Tosi, M. 2009, ARA&A, 47, 371Torrealba, G., Koposov, S. E., Belokurov, V., & Irwin, M. 2016,

MNRAS, 459, 2370

Torrealba, G., Belokurov, V., Koposov, S. E., et al. 2018, MN-RAS, 475, 5085

Udalski, A., Szymanski, M. K., & Szymanski, G. 2015, Acta As-

tron., 65, 1Valdes, F., Gruendl, R., & DES Project 2014, Astronomical Data

Analysis Software and Systems XXIII, 485, 379

Vargas, L. C., Geha, M., Kirby, E. N., & Simon, J. D. 2013, ApJ,767, 134

Venn, K. A., Shetrone, M. D., Irwin, M. J., et al. 2012, ApJ, 751,

102Voggel, K., Hilker, M., Baumgardt, H., et al. 2016, MNRAS, 460,

3384Walker, M. G., Mateo, M., & Olszewski, E. W. 2008, ApJ, 688,

L75

Walker, M. G., Mateo, M., Olszewski, E. W., Sen, B., &Woodroofe, M. 2009, AJ, 137, 3109

Walker, M. G., Mateo, M., Olszewski, E. W., et al. 2009, ApJ,

704, 1274Walker, M. G., Belokurov, V., Evans, N. W., et al. 2009, ApJ,

694, L144

Walker, M. G., Mateo, M., Olszewski, E. W., Bailey, III, J. I.,Koposov, S. E., Belokurov, V., & Evans, N. W. 2015, ApJ,

808, 108

Walker, M. G., Mateo, M., Olszewski, E. W., Koposov, S., Be-lokurov, V., Jethwa, P., Nidever, D. L., Bonnivard, V., Bailey,

III, J. I., Bell, E. F., & Loebman, S. R. 2016, ApJ, 819, 53

Walsh, S. M., Willman, B., & Jerjen, H. 2009, AJ, 137, 450Weisz, D. R., Koposov, S. E., Dolphin, A. E., et al. 2016, ApJ,

822, 32Williams, A. A., Belokurov, V., Casey, A. R., & Evans, N. W.

2017, MNRAS, 468, 2359

Willman, B., Dalcanton, J. J., Martinez-Delgado, D., et al. 2005,ApJ, 626, L85

Willman, B., Blanton, M. R., West, A. A., et al. 2005, AJ, 129,

2692Willman, B., Geha, M., Strader, J., et al. 2011, AJ, 142, 128

Willman, B., & Strader, J. 2012, AJ, 144, 76

Woosley, S. E., & Weaver, T. A. 1995, ApJS, 101, 181York, D. G., Adelman, J., Anderson, J. E., Jr., et al. 2000, AJ,

120, 1579

Zhu L., van de Ven G., Watkins L. L., Posti L., 2016, MNRAS,463, 1117

This paper has been typeset from a TEX/LATEX file prepared by

the author.

MNRAS 000, 1–21 (2018)